SociaHive Docs

Changelog

Notable changes to the SociaHive SDKs, MCP server, and REST API.

Public-facing changes that affect SDK consumers, MCP clients, or REST API callers. We don't log every internal refactor.

For version bumps that don't have a notable user-visible change, we publish the new version without a changelog entry — assume those are bugfix-only.

2026-11-21

Every tool from the start in Claude Code

  • Claude Code sessions, including Claude's cloud tasks, now get every tool when they connect, like every other client. Before, they started with the core tools and had to load other groups with enable_toolset, and cloud tasks never saw the loaded tools ("No such tool available").
  • ?toolsets=a,b (or the X-MCP-Toolsets header) still gives a shorter list: the core tools, those groups, list_toolsets and enable_toolset. See tool groups.

2026-11-20

More post options on every plan

  • These options now work on every plan. Before, they were dropped without an error unless the account was on an internal preview. See Post options.
    • x_reply_settings on an X post.
    • ig_collaborators on an Instagram post or reel.
    • contentType: "story" on Facebook.
    • yt_tags and yt_thumbnail_url on YouTube.
    • tiktok_is_aigc on TikTok.
  • X Articles (contentType: "article" on X) are available on the Studio, Pro and Agency plans, up to 20 each calendar month across all your X accounts. A new error code, ARTICLE_MONTHLY_LIMIT, means the month's Articles are used up. See Post options.
  • detach_post_automation is available to everyone, in the scheduler-advanced tool group.

2026-11-18

YouTube comment automations from MCP

  • New tool create_youtube_comment_flow creates a YouTube comment automation as a draft: it replies publicly to comments on your videos and Shorts, on every comment or only those with a keyword, on every video (new ones included) or the videos you name, with one of up to 10 reply texts in turn, after an optional delay and with optional per-video reply limits. Turn it on with activate_flow. It is in the flows-advanced tool group. See Tools reference.
  • update_node edits it on its trigger step, and duplicate_flow copies it.
  • get_capabilities (1.1.0) lists automation for YouTube accounts that can use comment automations, and the sociahive://trigger-types resource lists youtube_comment on YouTube.
  • Fixed: a plan change no longer turns off live YouTube comment automations, and an upgrade turns back on the ones it paused.

2026-11-17

First comments and thread parts on every plan

  • firstComment (Instagram, Facebook, LinkedIn, YouTube) and parts (X, Threads, Bluesky) on create_scheduled_post, and platform_overrides.<platform>.first_comment and .parts on the REST API, now work on every plan. Before, they were dropped without an error unless the account was on an internal preview. See Post options.
  • get_scheduled_post (1.2.0) returns each platform's followups: whether its first comment and each thread part was sent, failed (with the error) or may have posted.

TikTok comments, Duet and Stitch

  • allow_comment, allow_duet and allow_stitch ("true") on a TikTok target turn on comments, Duet and Stitch for that post. They were already read when the post publishes, but were not documented, so posts made through the API went out with all three off. privacy_level and the commercial disclosure keys are documented too. See Post options.

Plugs on YouTube

  • add_post_milestone_action (1.1.0) and POST /api/v1/posts/{id}/plugs take a YouTube target: when the video reaches a number of likes, comments or views, SociaHive adds your comment, once. YouTube is checked every 30 minutes for 48 hours.

2026-11-16

One Facebook Page connection

  • Connecting Facebook now connects the whole Page in one step: posts, and Messenger DMs and comments when your plan automates Messenger. The Messenger connection is gone from the connect window.
  • Underneath, a Page is still two accounts: facebook for posting and messenger for automations, with the same Page id. list_accounts and GET /api/v1/connected-accounts return both, as before. A Page counts once toward your plan's account limit.

Automation runs counted on every channel

  • Runs started by Facebook comments, Messenger, WhatsApp and Telegram now count in an automation's triggers, users and CTR. Before, only Instagram runs were counted.
  • The flow.triggered webhook now fires for those runs too.

2026-11-15

  • Link buttons in an automation's messages now open through a short SociaHive redirect, so a tap counts toward the automation's clicks and CTR. Instagram and Messenger don't report link taps, so before only reply buttons were counted.
  • Only buttons are tracked. Links typed inside a message's text are sent unchanged.

TikTok uploads over 10 MB and inbox captions

  • TikTok video uploads over 10 MB no longer fail with "The total chunk count is invalid".
  • A video sent to the TikTok inbox (tiktok_inbox_upload: "true") now carries the post's text as its caption. If TikTok doesn't accept a caption for the draft, the video is still sent without one.

Instagram trial reels on every plan

  • ig_trial_graduation (MANUAL | SS_PERFORMANCE) and ig_share_to_feed on an Instagram reel now work on every plan. Before, they were dropped without an error unless the account was on an internal preview. See Post options.
  • The create_scheduled_post tool no longer lists options that are not available on every plan (X reply settings, X Articles, Facebook stories, YouTube tags and thumbnail, TikTok AI label).

2026-11-11

Posts only go to your own accounts

  • Every post you create, edit, schedule or publish must name connected accounts you own. A target with another account's id, or an id that doesn't exist, is refused with 404 and code account_not_found (over MCP, a validation_error naming the account and listing your connected accounts). Nothing is saved.
  • At publish time the same check runs again, so a post can never go out through an account its owner doesn't hold.
  • The connect screen for AI apps now shows which SociaHive account you're signed in as, with a link to use another account.

Account handles match only where a tool can work

  • An accountId given as a handle (@name) now matches only accounts on platforms the tool can use. A handle you have on Instagram and on YouTube means the Instagram account for flow tools, since YouTube runs no flows.
  • list_flows takes accountIds (up to 10 ids or handles) and returns accountId on each flow. When listing, a handle that still matches several accounts lists the flows of all of them instead of failing; search and rank still ask which account.
  • When a handle can't be resolved, the error's data carries choices (value, label): the accounts the user may mean, for them to pick from.
  • No REST route or SDK method changed.

2026-11-10

@SociaHive in Slack

  • Pro and Agency accounts can add @SociaHive to Slack. People in the workspace ask about posts, automations and results, and get drafts back. Anything that would go live needs an approver on the SociaHive team to tap Confirm.
  • The account owner or a team admin adds it from Integrations and manages it in Settings > Connected apps: allowed channels, what it can do (read, draft or act), the default account, channel accounts, approvers, today's usage and uninstall.
  • Slack: mention it in a channel, message it directly, use the assistant panel, Ask SociaHive about this on a message, or /sociahive. See SociaHive in Slack.
  • Workspace limits: 300 replies a day on Pro and 2,000 on Agency, resetting at midnight UTC.
  • No REST route or SDK method changed.

Calendar for named periods

  • get_scheduler_calendar takes range: today, this_week, next_week, next_7_days, this_month or next_month, and the server works out the dates (UTC, weeks start on Monday), so "what's scheduled this week?" needs no dates from you. For exact dates, set range to custom and pass fromDate and toDate (a date-only toDate covers that whole day). A named range ignores any dates sent with it. Calls with only fromDate and toDate still work; with nothing, you get the next 7 days.

2026-11-07

Account and permission checks

  • /api/v1/account (delete, cancel deletion, deletion status) now works only from SociaHive or the SociaHive app. Other callers get 403 human_auth_required.
  • /api/v1/agent/chat and /api/v1/agent/confirm now apply the scopes your API key or connected app was granted, as the MCP server always has. A key or app without a tool's scope can't run that tool, and an app connected with read-only access can't make changes.
  • Flows created through the API (/api/v1/flows/generate, /api/v1/flows/from-template, flow import) use the connected account you name, and refuse with 404 an account that isn't yours. The account's platform id is taken from the account itself, so platform_user_id in the request is no longer used.
  • Turning on a flow whose connected account isn't yours returns 400 account_not_owned.
  • Banning an account, signing out everywhere, or requesting account deletion now also revokes its API keys and connected apps. Create new keys and reconnect your apps afterwards.

Webhooks

  • TikTok and Telegram events are now verified with each platform's own signature and delivered again. This also brings back inbox messages for connected Telegram bots.

2026-11-06

Approvals and post comments in the REST API

  • Every plan can now use approval steps, post comments and the review notifications, so their REST routes are in the API reference and both SDKs' types:
    • Approvals: GET /api/v1/approvals (posts waiting for your approval), GET /api/v1/approvals/{postId}, POST /api/v1/approvals/{postId}/decision, POST /api/v1/approvals/{postId}/request-changes and POST /api/v1/approvals/undo.
    • Comments: GET and POST /api/v1/comments/{postId}, POST /api/v1/comments/{postId}/{commentId}/resolve, and GET /api/v1/comments/inbox with POST /api/v1/comments/inbox/seen.
    • Notifications: GET and PATCH /api/v1/me/notification-matrix, and POST and DELETE /api/v1/me/notification-matrix/mutes.

X Article titles: 100 characters

  • x_article_title takes up to 100 characters (was 200). X refuses a longer title when the Article is published, so a longer one is now refused when the post is scheduled, with the reason, instead of failing at its time.
  • When X refuses an Article, the error message names the fields X pointed at.

2026-11-05

  • A WhatsApp tracked link can open WhatsApp with the bare code instead of REF_<code>. create_ref_url takes plainCode: true; the REST route takes plain_code: true. Existing links are unchanged. Available on every plan. See Send people between channels.

YouTube watch time in get_post_analytics

  • get_post_analytics (1.2.0) returns avgWatchTimeSeconds and avgViewPercentage (0 to 100, one decimal) on the row of a YouTube video, once YouTube Analytics has the numbers (usually 2 to 3 days after publishing). It needs the YouTube account to be reconnected once to grant read access to YouTube Analytics. Available on every plan.

Reels watch time in get_post_analytics

  • get_post_analytics (1.1.0) returns avgWatchTimeSeconds and totalWatchTimeSeconds on the row of an Instagram Reel, in seconds with one decimal, once Instagram has reported them. Other platforms and other Instagram posts leave both out.

Pause public replies on a busy post, or switch them off for one post

  • Comment trigger settings gain comment_reply_post_cap_per_day: at most this many public replies on one post in 24 hours (1 to 1000). The 24 hours start at the first counted reply, and replies resume by themselves when they end. null removes it.
  • comment_reply_disabled_media_ids lists posts (up to 100 media ids) where the trigger sends no public reply, without turning the replies off for the whole flow. DMs still go. null switches them back on.
  • Both apply to Instagram and YouTube comment triggers, and count or skip the comment reply and keyword replies alike.

Platform options in get_scheduled_post

  • Each platform result in get_scheduled_post now includes options (the platform options set on that target, such as li_document, li_document_title, yt_title or x_reply_settings) and contentType when the target has one, so you can check a post before it goes out.
  • The metadata description on create_scheduled_post and update_scheduled_post now says that li_document_title is the title text shown above the LinkedIn document (plain text, up to 400 characters; without it the first line of the text is used).

Tool groups

  • Tools now come in groups. Claude Code starts with the core group plus list_toolsets and enable_toolset, and loads another group when a request needs it. Other clients still get every tool.
  • Choose groups when you connect with ?toolsets=a,b on the server URL or the X-MCP-Toolsets header; ?toolsets=all gives every tool. A tool that is not in the list can still be called by name.
  • enable_toolset now names the groups your account can load and what each does, so an assistant finds a tool such as restore_scheduled_post or bulk_delete_scheduled_posts without loading groups first. Tool groups

2026-11-04

Fewer tools for the same lookups

  • list_flows takes a mode: search finds a flow by name (query); rank ranks flows by performance (metric: triggers, completions, unique_users or button_clicks). It also takes accountId. It replaces search_flows and list_flows_with_stats.
  • list_published_media takes a mode: latest returns the newest item, id one item by its platform media id (id), search caption search results (query). It replaces get_latest_published_media, get_published_media and search_published_media. Each mode returns { items }. id and query are read only in their own mode.
  • The replaced tools, and generate_flow_v1_0_0, still answer when called by name, so existing scripts keep working, but they are no longer in the tool list. CLI: sociahive flows list takes --query (search), --stats or --metric (rank) and --account-id.
  • create_ref_url now says what a link without flow_id opens: the account itself on platform (Instagram, the default: its profile; Messenger, Telegram or WhatsApp: a chat), counting each click. The name input is described as the owner's own label for the link. No change to what the tool does.

2026-11-03

  • A WhatsApp tracked link can open WhatsApp with your own text around the code ({code} marks where it goes; without it the code goes at the end; up to 300 characters). create_ref_url takes it as messageText. Available on every plan. See Send people between channels.

LinkedIn document posts

  • A LinkedIn post with two or more images can go out as one document (a swipeable PDF, one page per image): li_document: "true" and an optional li_document_title in the LinkedIn target's metadata. Available on every plan. See post options.

Edit a live flow without changing it for people until you publish

  • An edit to a live flow (published or paused) is now saved as unpublished changes, and people keep getting the live version. This covers add_node, update_node, delete_node, add_edge, update_edge, delete_edge and revert_flow_to_version; before, they refused a published flow. Their answer has pendingChanges: true and a notice.
  • New tools publish_flow_changes and discard_flow_changes (flows:write). A call with confirm: true and no expectedRev returns a preview (steps added, changed and removed, whether the trigger changed, people partway through a removed step, who made the changes) and a rev; a second call with confirm: true and expectedRev set to that rev publishes or discards. Publishing runs the same checks as turning a flow on, and a paused flow stays paused. Discarded changes can be restored in the SociaHive app while they are kept.
  • get_flow returns pendingChanges on a live flow, and version: "pending" returns the steps of the unpublished changes. list_flows, list_flows_with_stats and search_flows mark such flows with hasPendingChanges: true.
  • deactivate_flow on a flow with unpublished changes is refused (current: "pending_changes"): publish or discard them first.
  • REST: GET /api/v1/flows/{id}/pending (the changes and their rev), POST /api/v1/flows/{id}/pending/publish with { "expectedRev": "..." }, and POST /api/v1/flows/{id}/pending/discard. A client that sends X-SociaHive-Client-Capabilities: pending-flow-versions writes its edits to a live flow into the unpublished changes; without it, an edit to a live flow is checked against the draft rules, refused if it adds a problem that would block publishing, and otherwise applied live; it is also refused with 409 pending_changes_exist while the flow has unpublished changes. Codes: Error reference.
  • CLI: sociahive flows publish-changes <flow> [--yes] and sociahive flows discard-changes <flow> [--yes] print the changes, and --yes publishes or discards them.
  • Turning off public comment replies on a live flow, step by step: Comment trigger settings.

Upload several local files in one call

  • request_media_upload (1.2.0) takes files, a list of 1 to 25 { fileName, mimeType, sizeBytes }, and returns uploads (one uploadId, uploadUrl and headers per accepted file, in the order given), errors for the refused ones (index, fileName, code, reason, retriable), method and expiresAt once, maxBytes, and a summary with status ok or partial. A refused file never blocks the others; the plan check and the hourly cap refuse the whole call, and a batch that does not fit under the cap uses none of it.

  • finalize_media_upload (1.2.0) takes uploads, a list of 1 to 25 { uploadId, fileName? }, with one addToMediaLibrary for all of them, and returns items and errors in the same way. Each file is checked as before (size against the plan, type from its first bytes; a file that fails is deleted); an upload whose PUT has not landed yet is a not_found error with retriable: true, so it can be finished in a later call.

  • The single-file forms are unchanged. Every uploadUrl is signed for its own file's size and type, so each file goes to its own URL with its own headers, for example:

    # after request_media_upload with `files`, with its result in uploads.json;
    # the files are in $dir (fileName is a name, not a path). Every entry of
    # `headers` is sent as returned: the signature covers them.
    jq -c '.uploads[]' uploads.json | while read -r u; do
      args=()
      while IFS= read -r h; do args+=(-H "$h"); done \
        < <(jq -r '.headers | to_entries[] | "\(.key): \(.value)"' <<<"$u")
      curl --fail --upload-file "$dir/$(jq -r .fileName <<<"$u")" "${args[@]}" "$(jq -r .uploadUrl <<<"$u")"
    done

    then finalize_media_upload with uploads: [{ uploadId }, ...]. A 55-file day is 3 calls of each tool instead of 55.

2026-11-02

Turning on a flow and importing a file from a URL ask for approval

  • activate_flow and upload_media_from_url now ask for confirmation before they run, like publish_post_now: a flow that goes live starts messaging people, and an import fetches an address from outside SociaHive. The confirmation shows the flow and the account it messages from, or the whole URL with its query string.
  • With an API key (CLI, SDK, MCP with an sk_ key), pass confirm: true to run them in one call. sociahive flows activate <id> --yes confirms from the CLI.
  • confirm: true now works only with an API key. This changes today for apps connected with OAuth (Claude.ai, ChatGPT and other apps you sign in to): they no longer run a confirmed action in one call. Until SociaHive starts asking the person directly (below), they confirm with the token from the first call: tell the person what will happen, and call again with __confirmToken once they agree.
  • Confirmations now describe the action from its arguments (the text that would go out, when, on which accounts, every link in full) when the tool has no wording of its own.
  • From 2026-11-09 for admins and 2026-11-23 for everyone, in Hive AI and in apps connected with OAuth: once a tool that returns text written by other people (see below) has run in a conversation, SociaHive asks the person before any change. Apps connected with OAuth no longer receive a confirmation token: SociaHive asks the person in the app's own approval prompt when the app supports it, or returns a link to an approval page in SociaHive (code: "approval_required", approvalUrl). Calls made with an API key are not affected.

Resources need read scopes, and unverified apps are labelled

  • Reading an MCP resource now needs its read scope: sociahive://flows and sociahive://flows/{id} need flows:read, sociahive://posts/upcoming and sociahive://posts/{id} need posts:read, sociahive://accounts needs accounts:read, and sociahive://audit/recent and sociahive://health need settings:read. Without it the resource answers auth_error with missingScope. The reference resources (platforms, trigger types, docs, changelog) are unchanged.
  • An app that registered itself (dynamic client registration) is shown as "Unverified app" on the consent screen, with the website it will send you back to and the website it gave. Such apps are no longer granted agent:destructive, so they cannot call tools that delete. An app named after an AI vendor whose redirect_uris are all on that vendor's https domain (for example Claude on claude.ai, ChatGPT on chatgpt.com) is the vendor's own connector: it is not labelled and keeps agent:destructive. A localhost redirect does not count.
  • A registration named after an AI app or SociaHive (for example "Claude" or "ChatGPT") is refused unless every redirect_uris entry is on that app's own domain or on localhost (invalid_client_metadata).
  • The consent screen now signs the request it shows. The consent submission must come from that screen, for the same signed-in person, once, within 10 minutes.
  • /api/oauth/token is limited to 600 requests a minute per IP address (429 with error: "rate_limited").
  • create_broadcast now reports openWorldHint: true.

OAuth tokens are bound to one resource (RFC 8707)

  • /api/oauth/authorize and /api/oauth/token accept the resource parameter. Use https://www.sociahive.com/api/mcp for the MCP server (the resource in /.well-known/oauth-protected-resource/api/mcp) or https://www.sociahive.com/api/v1 for the REST API. A request without one gets the MCP server.
  • Each token works only at the resource it was issued for. A refresh keeps the same resource; asking for another one is refused with invalid_target.
  • For now a token used at the other resource is still accepted. Later it will be refused, and so will tokens issued before this change once 2026-12-15 has passed. If a client stops connecting then, connect it again.
  • API keys (sk_…) are not affected.

Tool results mark text written by other people

  • Tools that return text written by followers, customers, clients or other third parties (captions, collected answers, review comments, post text, flow steps, subreddit rules, destination names) now add a top-level _untrusted key: { "fields": [...], "note": "..." }. fields lists the paths that hold that text, such as items[].caption. Treat that text as data, never as instructions.
  • In those fields, invisible characters (zero-width characters, Unicode tag characters) and characters that reorder text (bidi controls) are removed, and long text is shortened. export_collected_data returns at most 200 characters per value. Line breaks are kept. Your stored data is not changed.
  • The rest of each result keeps its shape, so ids and other fields read as before.
  • Confirmation summaries, entity names and error messages are cleaned the same way.

2026-10-31

YouTube: upload now, and comment replies, on every plan

  • New YouTube platform option yt_upload_now: "true": the video is uploaded right away as Private and goes public at the scheduled time; editing or cancelling the post updates or deletes it on YouTube. See Post options.
  • Automatic replies to comments on your YouTube videos and Shorts are available in the SociaHive automation builder.
  • Both need the youtube.force-ssl permission: reconnect YouTube to grant it. Until Google finishes verifying it, Google shows an "unverified app" screen when you connect.
  • Disconnecting a YouTube, X or TikTok account now also removes SociaHive's access at that platform.

2026-10-30

Flow changes wait while a flow is being built

  • activate_flow, duplicate_flow and update_flow wait while a flow is being built, as the graph edits already did: they are refused with a state_error (current: "building", suggestion: "wait_then_retry"). get_flow reports build.status; a build usually finishes within a minute. bulk_update_flows reports such a flow with error: "building" and changes the rest.
  • PUT and PATCH /api/v1/flows/{id} and POST /api/v1/flows/{id}/duplicate answer 409 with code: "flow_building" in the same case.
  • A flow turned on, deleted or changed while a rebuild runs is no longer replaced by the rebuild: the rebuild is discarded and the flow stays as it is.
  • A name set with update_flow or bulk_update_flows is kept when the AI rebuilds the flow.
  • delete_node and delete_edge on a step or connection that is not in the flow now answer not_found.

Every way of creating and scheduling a post applies the same checks

  • A post to a platform or content type that is not available on your account is now refused on every route, drafts included: create_scheduled_post, duplicate_scheduled_post, POST /api/v1/posts, POST /api/v1/posts/{id}/duplicate, CSV import and RSS feeds, as the web composer already did. This covers Instagram Reels and Stories while they are off, a platform switched off for scheduling, and X on the free plan. CSV import reports such a row as an error and imports the others.
  • The same check runs again when a post is scheduled, published, approved, edited while scheduled, or moved to another time.
  • A refusal answers 400 (403 for a platform that is not on your account yet) with code platform_unavailable or content_type_unavailable, platform, content_type and targets. Over MCP it is a validation_error on platforms or platforms.contentType.
  • create_scheduled_post with scheduledAt now leaves nothing behind when the post cannot be scheduled, as POST /api/v1/posts does: the draft is removed and the reason is returned. Before, the draft stayed.
  • create_scheduled_post with scheduledAt and schedule_post now count against your plan's monthly scheduled posts, as the web app and the REST API do.
  • POST /api/v1/posts with a scheduled_at in the past is now refused (Scheduled time must be in the future) instead of saving a draft.
  • schedule_post with attach_automation: when scheduling is refused, the reason is returned even if taking the automation off again fails. Turn the automation on with arm_post_automation or on the post page.
  • request_media_upload, finalize_media_upload, upload_media_from_url, POST /api/v1/uploads/presign and sociahive upload accept PDFs (application/pdf), checked by the file's first bytes. A PDF is for a file block in a flow message (a WhatsApp document, a Messenger or Telegram file); the upload tools return mediaType: "document" for it. A PDF is refused as post media (create_scheduled_post, update_scheduled_post, POST /api/v1/posts). See Sending a file.
  • A message step with a file, document, image, video or audio block that has no file is refused when the flow is turned on, on WhatsApp, Messenger and Telegram steps too. Before, such a WhatsApp block went live and sent nothing.
  • create_ref_url and POST /api/growth-tools/ref-urls refuse a WhatsApp link when the WhatsApp number it would open is not connected (a conflict error, HTTP 409 whatsapp_not_connected). A WhatsApp link whose number was disconnected later now shows "This link is not available right now" instead of opening your Instagram profile.
  • A WhatsApp link sent to a known contact now links that person's WhatsApp number to the same contact, and a WhatsApp number that matches one contact's phone links too. A link opened more than a day after it was sent only suggests the match, since it may have been forwarded. See The same person on two channels.

Comment reply texts take turns, and per-post caps are checked

  • A comment trigger with several public reply texts (comment_reply_templates) now uses them in turn on each post, instead of picking one at random, so the same text no longer shows twice in a row under one post. Each trigger keeps its own turn on each post.
  • A text Instagram would refuse (too long) is no longer picked, so it can no longer make a reply fail while the trigger has other texts.
  • comment_reply_post_cap_per_hour and comment_reply_post_cap_total outside 1 to 1000, or not whole numbers, are now refused with a validation_error on update_node and add_node, as the REST API already did. Before, such a value was saved and silently ignored.
  • Editing a comment trigger in the flow editor no longer removes the per-post caps (or any other setting the editor does not show) set over MCP or the API. The editor now has a "Limit public replies on each post" field.
  • Every comment trigger setting is listed in the tools reference.

generate_flow keeps to what you asked

  • A flow no longer asks for the same email, phone number or answer twice. When a step asks again for something the person was already asked for on every way to that step, the repeat (and a thank-you message right after it) is removed before the flow is saved.
  • In an English description, an email or phone question you did not ask for is left out: a request that names your email gets no phone question, and a plain "send them my link" gets no question at all. A request that asks for leads without naming a field keeps the question the flow needs.
  • A wait of 15 minutes or more followed only by more messages, tags or emails is left out unless the description asks for one ("follow up", "remind them", "2 days later", "tomorrow"). Mentioning a time, as in "we reply within 24 hours", no longer counts as asking for a follow-up.
  • With allowFollowGate: true on an Instagram flow, the flow always has a real Ask-to-follow step. If the draft has none, the steps after the comment reply go behind one, and a message whose "I followed" button only imitated the check is replaced by it.

Connect, re-point and remove connections by button

  • get_flow lists every step's outputs: each button, quick reply, card button and branch with its handle, its label and connectedTo, and the plain next output (no handle). Each entry in edges has its id, and a connection stored with source_handle shows its output too.
  • add_edge takes the output as a handle or as its label (a button title or branch name, matched without regard to case). The outputs it accepts are the ones the flow editor draws; an unknown output is refused with the list, for example Valid outputs of Choose: button-0 (Yes, please), button-1 (No, thanks). A step whose buttons route has no plain next output.
  • A button or branch carries one connection. Connecting one that is already connected is refused; use the new update_edge.
  • New: update_edge sends a connection to another step (newTarget). Name it by edgeId, or by source and sourceHandle, plus target when the plain output has several connections. Like the other graph edits it works on drafts, refuses routing out of a payment or Ask-to-follow step, and saves a version first.
  • delete_edge accepts the same source and sourceHandle (and target) as well as edgeId.
  • CLI: sociahive flows update-edge and sociahive flows delete-edge.

WhatsApp buttons in more than one message

  • A WhatsApp step with buttons in two messages sent a tap on the second message's buttons along the first message's connections. Buttons are now numbered across the whole step, as the flow editor numbers them, so each button follows its own connection. Buttons sent before this change keep working.

A branch with nothing connected ends there

  • When a condition, randomizer or payment step's chosen branch had no connection, the flow sometimes went on with every step after it. It now ends there for that person, as a branch with no connection always does.

YouTube title, visibility, audience and category

  • New YouTube platform options on every plan: yt_title (up to 100 characters), yt_privacy (public, unlisted, private), yt_made_for_kids ("true", "false") and yt_category (a YouTube category id). Set them in platforms[].metadata on POST /api/v1/posts and PATCH /api/v1/posts/{id}, or in a target's metadata on create_scheduled_post and update_scheduled_post. See Post options.
  • Scheduling, publishing or approving a post now checks the YouTube title and description: a title is required, at most 100 characters, no < or > in the title or description, and a description of at most 5,000 bytes. A post that breaks them is refused with TEXT_TOO_LONG and a sentence for the YouTube target, instead of failing at publish time. Drafts still save.
  • A YouTube upload that may have created the video is no longer retried (a retry made a second video). SociaHive looks for the video on the channel; if it is not found, the target fails with AMBIGUOUS_MAY_HAVE_POSTED.

2026-10-29

Upload a local or private file from your AI tool

  • New request_media_upload and finalize_media_upload, for AI tools that can run code (Claude Code, Cursor, scripts). request_media_upload takes fileName, mimeType and sizeBytes and returns an uploadUrl with the headers to send; the tool sends the file there with an HTTP PUT; finalize_media_upload with the uploadId checks the stored file and returns { url, key, sizeBytes, mimeType, mediaType }, the same as upload_media_from_url, ready for create_scheduled_post. This is how a file on your computer, or one in a Google Drive folder that is not shared publicly, gets into SociaHive without an API key. JPEG, PNG, WebP, MP4 and MOV up to your plan's upload size; the upload URL expires 15 minutes after it is issued. finalize_media_upload reads the first bytes of the file and deletes it if they are not the type it was uploaded as, and adds the file to your media library unless addToMediaLibrary is false. Both need the uploads:write scope.

Instagram Reels and Stories for every plan

  • Scheduling Instagram Reels (content_type: 'reels') and Stories (content_type: 'stories') is available on every plan, from the app, the REST API and create_scheduled_post.

Cap public comment replies per post

  • A comment trigger can cap its public replies on each post: comment_reply_post_cap_per_hour (in a clock hour) and comment_reply_post_cap_total (overall), each 1 to 1000. Set them in the trigger's config with update_node; set one to null to remove it. Over the cap, the public reply is skipped and the DM still goes, except on a trigger that only replies publicly (comment_reply_only), where nothing is sent. The counter is per post and shared by every flow that replies on that post. It counts the comment reply and keyword replies.

Turn on a post's automation from your AI app

  • New arm_post_automation: turns on "Turn on when this post goes live" for the comment automation attached to a scheduled post. It takes two calls. A call with confirm: true and no contentHash turns nothing on and returns the preview: the automation's steps, its audience and a contentHash. After the person has seen the preview and agreed, a second call with confirm: true and that contentHash turns it on. A contentHash that no longer matches the automation is refused. The same checks as the post page apply, and the account owner is notified.
  • CLI: sociahive posts arm <id> prints the automation's steps; add --yes to turn it on. Also sociahive posts restore <id> and sociahive flows restore <id-or-name>.
  • A WhatsApp ref link now starts the flow it was made for, even when another flow also answers ref links.
  • A WhatsApp ref link made with create_ref_url opens the number of its flow's account, not just the first connected WhatsApp number.
  • The pre-filled REF_ text is now case-sensitive.

Editing a live flow

  • Editing a published flow still returns state_error; its message now names both ways to change it: deactivate_flow, edit, activate_flow, or duplicate_flow and edit the copy. See Flow patterns.

Delete and restore for every plan

  • delete_flow, delete_scheduled_post and bulk_delete_scheduled_posts are available on every plan. Each delete runs only after a confirmation. A connected AI app can delete up to 50 items a day, at most 10 deletes in 10 minutes, and the account owner is notified.
  • New: restore_flow brings a deleted flow back from the trash as a draft (30 days). restore_scheduled_post brings a deleted post back; a post that was scheduled comes back as a draft to schedule again.
  • disconnect_account stays unavailable to AI apps.
  • AI apps can now be granted the broadcast, sequence, growth-tool and tag permissions (broadcasts:write, sequences:write, growth_tools:write, tags:read, tags:write), and agent:destructive for deletes. Reconnect the app to pick them up.

Instagram accounts that need reconnecting to post

  • When Instagram refuses a post for an Instagram account connected before our current sign-in, the failed platform result now reads "Instagram didn't let SociaHive post for @handle. Reconnect it in Connected accounts, then try again." (code RECONNECT_TO_PUBLISH, from get_scheduled_post, GET /api/v1/posts/{id} and the post.failed webhook).
  • list_accounts returns reconnectToPublish: true on such an account, and GET /api/v1/connected-accounts returns reconnect_to_publish: true. The account stays connected and its automations keep running; reconnecting it clears the field.

Flows build only what you ask for

  • generate_flow no longer ends now and then without building a flow (the build stopped after noting the business details). Before, about 1 build in 10 failed this way.
  • generate_flow and generate_multiple_flows no longer add steps the request did not name. Asking for an email no longer adds a wait, a second message and a tag; a wait, a follow-up or a tag is added only when the request asks for it (a "funnel" still gets its follow-up).
  • The same field asked twice in a row (for example two email questions) is caught and fixed while the flow is built.
  • Ask-to-follow steps: set allowFollowGate: true on generate_flow or generate_multiple_flows, or allow_follow_gate: true on POST /api/v1/flows, when the request asks people to follow first. The request can be in any language. A warnings entry says when the flow is not on Instagram, where the step is not available.

Connections from buttons and the follow step

  • add_edge saves sourceHandle (and targetHandle). Before, the output was dropped, so a connection from a button (button-0) or from an Ask-to-follow step (followed, not_followed) left from the default output. Connections made over MCP before this fix: remove them with delete_edge and add them again.
  • add_edge refuses a step that is not in the flow and an output the step does not have, with the outputs it does have. Steps can be named by id or name.
  • The WhatsApp step summary names files, images and videos.
  • Facebook comment replies skip blank reply texts.

Build state on get_flow, and edits wait for the build

  • get_flow returns build: status is building, needs_clarification, failed or ready, with error or question when there is one. While a build runs, problems is empty.
  • Editing a flow with add_node, update_node, delete_node, add_edge, delete_edge or revert_flow_to_version while its build is running returns state_error with current: "building" and suggestion: "wait_then_retry". Before, the edit could be overwritten when the build finished.
  • A build that finishes after the flow was changed in the editor no longer replaces those changes. It ends with error.reason edited_during_build, and the changes are kept.

Fixes for AI clients

  • create_ref_url returns url as the full link to share (https://www.sociahive.com/r/…; it was a relative /ref/… path) and platform. With flow_id, the link opens that flow's channel (a WhatsApp referral flow gets a WhatsApp link). platform must be one of instagram, facebook, messenger, whatsapp, tiktok, telegram, twitter or linkedin, and flow_id must be one of your flows.
  • update_node with text on a comment, sms or email step changes its text. Before, it was refused.
  • On a comment trigger, the public reply goes out after the opening DM; comment_reply_timing is stored but does not change this. With comment_reply_only: true only the public reply is sent, at once.
  • A WhatsApp file block sends the document with its fileName. A block typed document is sent as a file; before, it was skipped.
  • create_flow takes the account by @handle or id and reads the platform from it. platformUserId is no longer needed. A flow for a WhatsApp account no longer defaults to Instagram.
  • list_capabilities lists only the tools this connection can call: the same list as the MCP tool list, scopes included.
  • schedule_post with attach_automation: when the post cannot be scheduled, the automation is removed again, so nothing is left attached to an unscheduled post.
  • upload_media_from_url accepts a Google Drive link to a file shared as "Anyone with the link". A link that opens a web page gets an error saying so.

2026-10-28

Tool descriptions describe, they don't instruct

  • Every tool description, input field description and the server instructions now state what a tool does, what each field means and what it returns. Wording that told the model when or how to call a tool is gone, so ChatGPT no longer shows "InstructionAttempt" on the permission dialog. Refresh the SociaHive app in ChatGPT once to pick up the new tool list.
  • Tool names, inputs, outputs and behaviour are unchanged.

Clearer tool choice for AI clients (server 1.1.0)

  • Descriptions of the most-used tools say when to use each one and which related tool covers the other cases ("Use this when… Do not use for…"), and that accounts, flows and posts are resolved from names and @handles, so no lookup call is needed first.
  • current_page_context is no longer listed over MCP: it only applies inside the SociaHive app. create_automation points to generate_flow, which builds the same draft and takes an account name or @handle.
  • The server reports version 1.1.0, so clients can tell the tool list changed.
  • Descriptions state what a tool does and leave risk to the tool annotations (readOnlyHint, destructiveHint): wording such as "nothing is sent" or "always safe", and hints about when to call or retry, are gone.

2026-10-27

Contacts in webhooks and Zapier

  • Webhook events about a person (flow.triggered, flow.error, newsletter subscriber events, data.collected) now carry contact: email, phone, tags and custom fields by name. See webhooks.
  • GET /api/v1/contacts takes platform_user_id, an exact match on one of the contact's platform ids, such as the senderId of flow.triggered.
  • The Zapier app has a new search, Find Contact: by contact ID, by the Sender ID of Automation Triggered, or by email, name or username. It needs the contacts:read scope.

2026-10-26

Plug and posting slot tools renamed

Clearer names for MCP clients, and one tool per operation:

  • list_plugs is now list_post_milestone_actions, and arm_plug is now add_post_milestone_action. Inputs are unchanged.
  • manage_posting_slots is replaced by list_posting_slots, add_posting_slot, update_posting_slot and delete_posting_slot. Each takes the fields its old action did, without action.
  • find_next_slot no longer holds the slot it returns, and its answer no longer has reservedUntil. To schedule into the next free slot, use create_scheduled_post with nextSlot: true.
  • upload_media_from_url, create_review_link and add_post_milestone_action now report openWorldHint: true.

The CLI commands (sociahive plugs, sociahive slots) are unchanged.

Plainer generate_flow description

The description MCP clients see for generate_flow now only says what the tool does, what it takes and what it returns. Some clients flagged the previous text because it told the model how to behave. The tool's inputs and behaviour are unchanged.

generate_flow takes a name

Pass name (2 to 5 words) and the new draft shows in the user's flows list under that name right away, marked "Building…" until its steps are ready. Without it, SociaHive names the draft from the description.

Every accountId takes a handle

Any tool that acts on a connected account now takes accountId as the account's @handle as well as its id, including create_scheduled_post (platforms[].accountId), list_scheduled_posts, arm_plug, list_account_destinations, get_reddit_post_rules, list_pinterest_boards and the automation accountId on schedule_post and update_scheduled_post. A handle must name one of the user's own accounts, on the target's platform when there is one. Otherwise the call returns validation_error on that field listing the choices, and nothing changes. A platform name ("instagram") still means the user's one account on that platform. No list_accounts call is needed first.

Accounts the user names are checked before anything runs

  • When accountId is left out and the request names an account ("On my latest post on @acme, ..."), generate_flow uses that account if it is connected. If it isn't connected, or it is a different account from the one already selected, nothing is built: the call returns validation_error on accountId listing the connected accounts, so the client can ask the user and retry with the accountId they choose.
  • generate_flow, search_flows and the published-media tools (list_published_media, get_latest_published_media, get_published_media, search_published_media) describe accountId the same way: the @handle the user named, or an id from list_accounts.
  • Every tool that acts on one connected account returns actedOnAccount (id, platform, username), the account it actually used.
  • The server now sends MCP instructions at initialize: pass a named @handle as accountId, ask the user when an account isn't connected, and report actedOnAccount.

2026-10-25

Name an account by its handle

generate_flow, search_flows and the published-media tools take accountId as an id from list_accounts or the handle the user typed (@health_wellness_dogo, health wellness dogo or a profile link), matched against the user's own connected accounts only. When accountId is left out and only one connected account fits, that account is used. When several fit, or the name does not match one exactly, the tool changes nothing and returns validation_error on accountId listing the accounts to choose from (closest matches first), so the client can retry or ask the user. Before, these calls failed with "Authentication failed". Tools reference

List the posts of one connected account

list_scheduled_posts takes an optional accountId (from list_accounts) and returns only posts that went, or will go, to that account. Each link in its results, and each result of get_scheduled_post, now carries the accountId it belongs to, so a workspace with two X accounts can tell their posts apart. Tools reference

Account errors say what went wrong

Errors about the account a tool should use (no account picked, an account that is not on the workspace, an account a tool cannot read) now carry a message saying what to do, for example that published media is available only for Instagram Business and Creator accounts. Before, MCP clients saw only "Authentication failed".

AI tools can build an "Ask to follow" step

generate_flow and add_node can now build "Ask to follow" on Instagram flows: the flow continues only once Instagram confirms the person follows the account. It is added only when the user asked for a follow check, never on the assistant's own initiative. Over MCP and the API, pass the new allowFollowGate: true; on generate_flow the description must also say that people must follow first. add_node accepts type: "follow_gate". The step's messages are the product defaults, and the user can change them in the flow editor. Calls without allowFollowGate work as before. Tools reference

add_node no longer accepts type: "facebook"

facebook was never a step type: it made a step the flow editor could not show and SociaHive could not send. Facebook Page automations use type: "messenger", as before. Tools reference

Images from ChatGPT go straight into posts

In ChatGPT, upload_media_from_url now also takes file: an image ChatGPT generated or a photo the user added to the chat. ChatGPT passes the file itself (the tool declares _meta["openai/fileParams"]), SociaHive copies it into the user's uploads, and the returned url is used as post media as before. Same limits as a URL import: JPEG, PNG, WebP, MP4 and MOV up to the plan's upload size. Send either url or file. The tool is now version 1.1.0; calls with url work unchanged. Tools reference

2026-10-24

Open the chat widget from your own button

Add data-sociahive-open="chat" to any button or link on your site and clicking it opens the chat widget. help, messages and home open those screens instead. window.SociaHiveWidget.open() takes the same values. Chat widget

2026-10-20

SDKs, CLI and the n8n node released

  • Node SDK @sociahive/sdk 0.2.0: npm install @sociahive/sdk. SDKs
  • Python SDK sociahive 0.2.0: pip install sociahive.
  • CLI sociahive 0.3.0: npx sociahive --help. It covers the scheduler features opened on 2026-10-19 (signatures, posting slots, plugs, client review and invite links, the media library and more).
  • Setup wizard @sociahive/mcp 0.2.0: npx @sociahive/mcp add connects SociaHive to your AI tool.
  • n8n community node n8n-nodes-sociahive 0.2.0: install it from Settings → Community nodes in n8n. n8n

Slack, Tumblr, Dribbble, Twitch and Kick

Five more platforms you can connect and schedule to on every plan: Slack channels, Tumblr blogs, Dribbble shots, and Twitch and Kick channel chat. They are in the platform values of the tools and the REST API, and use destinations, platform options and articles where the platform needs them.

2026-10-19

Scheduler features for every plan, and 8 more platforms

The scheduler features from the 2026-10-15 admin preview are open to these plans:

A call your plan does not include still answers 403 with code: "feature_disabled".

New platforms you can connect and schedule to on every plan: Telegram channels, Discord, Mastodon, Lemmy, Dev.to, Hashnode, WordPress and Listmonk. They are in the platform values of the tools and the REST API, and use destinations, platform options and articles where the platform needs them.

Pinterest boards, and Instagram Stories and Reels

  • Pins can be scheduled again. Every Pin is saved to a board, and nothing could choose one, so Pinterest posts failed with MISSING_BOARD.
    • The web composer now has a board picker.
    • New tool list_pinterest_boards returns each Pinterest account's boards; pass the chosen id as metadata.boardId on the Pinterest target of create_scheduled_post.
    • An account with a single board needs no boardId: the Pin is saved there.
  • Instagram Stories and Reels are in beta. A post that uses them is now refused with FEATURE_DISABLED when it is scheduled or published, from every surface, unless your account has the beta. The web app already refused them; create_scheduled_post, the REST API and CSV import did not. Drafts still save.

2026-10-15

Admin preview: scheduler features

These are on for admin accounts only for now. Every other account gets 403 with code: "feature_disabled" from the new endpoints and tools, and the new post fields have no effect. Each feature opens to more plans separately.

  • AI images for posts: POST / GET /api/v1/media/generate-image, tool generate_post_image, CLI sociahive media generate. Counted against a monthly allowance per plan. AI images
  • Media library: GET /api/v1/media-library, tool list_media, CLI sociahive media list. Media library
  • Short links and link clicks: links become /l/<code> at publish; GET /api/v1/posts/{id}/link-clicks, tool get_post_link_clicks, CLI sociahive posts clicks <id>. Short links
  • Signatures: /api/v1/signatures, tools list_signatures and create_signature, platform_overrides.<platform>.signature_id, CLI sociahive signatures list|add and posts create --signature. Signatures
  • Posting slots: /api/v1/posting-slots and GET /api/v1/find-slot, tools find_next_slot and manage_posting_slots, create_scheduled_post { nextSlot: true }, CLI sociahive slots list|add|remove|next and posts create --next-slot. Posting slots
  • RSS feeds written by AI: /api/v1/rss-feeds, tools list_rss_feeds and update_rss_feed, CLI sociahive rss list|pause|resume. RSS feeds
  • Alt text, video covers, separate edits, X long posts: alt_text, cover_url, cover_time_ms, platform_overrides.<platform>.detached on the existing post calls, text_limit on connected accounts; CLI --alt, --video-cover, --detached. Post options
  • Duplicate a post: POST /api/v1/posts/{id}/duplicate, tool duplicate_scheduled_post, CLI sociahive posts duplicate <id>. Duplicate, preview and stats
  • Client review links: /api/v1/posts/{id}/review-links, tools create_review_link (asks to confirm) and list_review_links, CLI sociahive posts review-link <id>, webhook event post.review_decision. Client review links
  • Client invite links: /api/v1/client-invites, tools create_client_invite (asks to confirm) and list_client_invites, CLI sociahive clients invite <group> and clients invites. Client invite links
  • Plugs: /api/v1/posts/{id}/plugs, tools list_plugs and arm_plug, CLI sociahive plugs list|add, webhook event post.plug_fired. Plugs

The tools reference lists every new tool with its scopes. The Node and Python SDKs 0.2.0 and the n8n node cover these endpoints (released 2026-10-20, see above).

2026-10-14

A caption and media per platform, and a fix for platform_overrides

  • Creating a post with platform_overrides (in POST /api/v1/posts, and in the web composer) failed with 400 "Maximum call stack size exceeded", even for a text-only override. Fixed. The created post now also comes back with its platform_overrides.
  • create_scheduled_post takes text and media on each platforms[] item: that platform's own caption and images or video, used instead of the post's. They are stored as the post's platform_overrides, next to parts and firstComment. Each platform's own limits apply (X 280 characters and at most 4 images, Instagram feed 4:5 to 1.91:1). update_scheduled_post does not take them yet.
  • CLI: sociahive posts create --platforms passes the same text and media per target, for example [{"platform":"twitter","text":"Short version","media":[{"type":"image","url":"..."}]}].
  • An unexpected server error on creating, updating, scheduling or publishing a post is now 500 with a generic message ("Failed to create post"). It used to be a 400 carrying the internal error text. Requests the server refuses still answer 400 with the reason.

X Articles (rolling out)

Long-form X Articles can be written and scheduled like any other post. Articles are being rolled out gradually; until your account has them, an Article is refused with FEATURE_DISABLED and never posted as a tweet.

  • An Article is an X target with content_type: "article" (contentType in create_scheduled_post). It must be the post's only target, has no X-specific text or thread parts, and does not repeat.
  • The post's text is the article body, in Markdown by default. Its images go in the post's media first, and the body refers to them by URL. Do not repeat the title in the body.
  • Target metadata: x_article_title (required, one line, up to 200 characters), x_article_format (markdown or content_state, which is X's raw block JSON with image data { media_url }) and x_article_cover_url (the URL of one of the post's images).
  • The post text limit on create_scheduled_post and update_scheduled_post is now 100,000 characters. Each platform's own limit still applies when the post is scheduled or published.
  • update_scheduled_post accepts media, which replaces the post's media list. This also makes sociahive posts update --media work.
  • A refused Article (ARTICLE_INVALID), an option the account cannot use (OPTION_UNAVAILABLE) and a feature not on the account yet (FEATURE_DISABLED) are now validation errors with the full reason, like a text that is too long. They used to come back as server errors.
  • If X does not confirm an Article was published (ARTICLE_PUBLISH_UNCONFIRMED), it may already be live. Retrying it through the API returns 409 ARTICLE_UNCONFIRMED unless the request sends acknowledge_unconfirmed: true (publish, schedule and update). publish_post_now never retries it; check X, then retry from the SociaHive web app.
  • Published posts now come back with their link wherever the platform allows one: url in get_scheduled_post results and the list and calendar links (X, Facebook, LinkedIn, YouTube, Pinterest and Bluesky, not only X), and permalink on each target in the v1 GET /posts and GET /posts/{id}. Instagram, Threads and TikTok links come only when the platform returned one at publish.
  • Saving a post with an image edited in the media editor (original_url, editor_state) no longer returns 400.
  • list_scheduled_posts, get_scheduled_post and get_scheduler_calendar return articleTitle for an Article, and its textPreview is the opening of the body as plain text. Posts can be found by an Article's title (postName).
  • CLI: sociahive posts create and posts update take --text-file <path> instead of --text. posts create also takes --content-type, and --title and --cover for an Article on the X target.
  • The sociahive://platforms resource lists the Article limits under twitter.article. The tweet limits are unchanged.

2026-10-13

URL import errors, and post statuses for older iOS builds

  • POST /api/v1/uploads/from-url and upload_media_from_url no longer answer 502 when the remote site is the problem (our CDN replaced that body with a bare "error code: 502"). They answer 422 with upstream_refused (the site answered an error; details.reason says e.g. "the site answered HTTP 403"), upstream_timeout or upstream_unreachable. Downloads now send User-Agent: SociaHive-MediaBot/1.0 (+https://www.sociahive.com), which sites such as Wikimedia require.
  • Clients can send X-SociaHive-Client-Capabilities (comma-separated). The SociaHive iOS app sends reminder-status; an iOS build that does not gets post status reminder_sent as draft and target status reminder_sent as pending. Other callers always get the real status.

Uploading media from the CLI and from a URL

  • New CLI command sociahive upload <file> [<file> ...] runs presign, PUT and verify for each file and prints [{ file, url, key, size_bytes }]. sociahive posts create --media-file <path> (repeatable) uploads first and adds the files as media. Needs uploads:write.
  • New tool upload_media_from_url and route POST /api/v1/uploads/from-url ({ url } → { url, key, size_bytes, mime_type }, uploads:write): copy a public image or video into your uploads, for clients that cannot send file bytes. Files up to 100 MB (or your plan's upload limit, if lower); upload larger files with presign. Behind the MEDIA_UPLOAD_FROM_URL flag (403 feature_disabled while off).
  • The presign response's required_headers is now documented: the PUT must send every entry exactly as returned, or storage answers 403. The response's key is now in the OpenAPI spec.

Scheduling and posts

  • POST /api/v1/posts with schedule_type: "scheduled" and a future scheduled_at now schedules the post in the same call (it used to save a draft). A refusal (text too long for a platform, an option the account cannot use) answers the same 400 as POST /posts/{id}/schedule, and no draft is left behind. Both now apply the plan's monthly scheduling cap (403 upgrade_required).
  • A create's response returns each target's metadata (it was {}).
  • A published TikTok post now carries its public post id and link (url in get_scheduled_post, links in list_scheduled_posts) once TikTok has them. A private post, or one still in TikTok's review, has no link; it is filled in later when TikTok publishes it, for up to 24 hours.
  • New post and target status reminder_sent ("remind me to post", rolling out behind a flag): the owner was reminded and nothing was posted by us.

More connected accounts, and a limit on automated accounts

Every plan can connect more accounts, on any supported platform in any mix: Free 3, Creator 10, Studio 20, Pro 30, Agency 150. Connecting refuses with account_limit_reached only past these new numbers.

A separate limit caps how many of those accounts can run automation: Free 0, Creator 1, Studio 4, Pro 10, Agency 100. An account counts once it has a live flow or AI Replies / AI Comments on. Turning on automation for one more account past the limit (activate_flow, bulk publish, starter kits, switching on AI Replies or AI Comments) returns automated_account_limit; accounts that already run automation keep running. The limit is being rolled out gradually and is not checked on every account yet.

2026-10-11

MCP tools carry titles and hints

Every MCP tool now has a title and explicit readOnlyHint, destructiveHint, idempotentHint and openWorldHint annotations, so clients such as ChatGPT and Claude can tell reads from writes and ask before anything that deletes data or goes public. The internal __cache_anchor tool is no longer listed over MCP.

When the account's plan doesn't include connecting AI apps, tool calls now return a plain "not included in this account's plan" message without a link.

2026-10-09

Outbound webhooks are available

Webhooks can now be set up by every account, from the dashboard (Integrations → Webhooks) or with /api/v1/webhooks. Events are delivered while the plan includes API access (the trial and paid plans). See Webhooks for the events, the payload and how to verify the signature.

Changes for anyone who used /api/v1/webhooks while it was limited to admin accounts:

  • events accepts the same event names as before except two internal ones, inbox.message and post.automation_attention, which are now refused and no longer delivered. Sent today: post.published, post.failed, flow.triggered, flow.error, whatsapp.template_status. Accepted but not sent yet: message.received, comment.received, flow.completed, data.collected, account.disconnected.
  • integration_type is unchanged (webhook, custom, n8n, make, zapier). The plan limit counts webhook, custom and n8n only; make and zapier keep the previous cap of 10 active subscriptions.
  • GET lists webhooks that are switched off too. PATCH accepts active; true switches a webhook back on and resets its failure count.
  • DELETE removes the webhook for good, so the same URL can be added again.
  • New optional account_ids limits a webhook to events about those connected accounts.
  • target_url must be a public HTTPS address; private and internal addresses are refused with 400.
  • Creating a webhook over your plan's limit returns 402 with code: "webhook_limit".
  • Responses include last_delivery_at, last_status and last_error.
  • flow.triggered now carries accountId and the account's real platform (it always said instagram before). whatsapp.template_status now carries accountId.

2026-10-06

New tool: export_collected_data

Returns the actual values collected by your flows (emails, phone numbers, names, answers), newest first, so an assistant can export or sync your leads. Inputs: flowId (optional; omit for all your flows), fieldType, since (ISO date), limit (1-500, default 100) and cursor. Output: items (id, flowId, flowName, fieldType, fieldLabel, value, valid, collectedAt), nextCursor and total. Needs the flows:read scope. Rolling out to admin accounts first; until it reaches yours, calls are refused.

get_collected_data hides email- and phone-shaped previews

get_collected_data stays a summary. Its preview field is now also left out when the value looks like an email or a phone number, whatever the field type. Use export_collected_data for the values.

2026-10-05

When an MCP client asks for any write scope, the SociaHive consent screen now offers a "Read-only access" box. If the person ticks it, the authorization code and every token minted from it carry only the requested *:read scopes plus agent:execute; all *:write scopes and agent:destructive are dropped. Write tools are then refused for the missing scope. Read the granted set from the scope field of the token response rather than assuming you got what you asked for.

Token revocation endpoint

POST /api/oauth/revoke implements RFC 7009. Send token (access or refresh, form-encoded or JSON), and optionally token_type_hint and client_id. Revoking either token revokes the pair. The response is 200 for any token, including unknown or already revoked ones; 400 invalid_request only when token is missing. The endpoint is advertised as revocation_endpoint in /.well-known/oauth-authorization-server, with revocation_endpoint_auth_methods_supported: ["none"].

People can also revoke a connected app from Settings → Connected apps. Either way the next request with the revoked token gets 401 invalid_token.

2026-10-02

generate_flow: iterateOnFlowId no longer rebuilds a flow a person has edited

generate_flow with iterateOnFlowId replaces every step of the target flow. It now works only on an empty draft, or on a flow whose content is still exactly what generate_flow last built. On any other flow the call fails with state_error and suggestion: "use_edit_tools"; nothing is changed.

To change one thing in an existing flow, use update_node, add_node, delete_node or add_edge.

Flows built before this date count as edited, so they can be changed with the edit tools but not regenerated in place.

get_flow returns the flow's steps

get_flow used to return only node and edge counts. It now also returns nodes (each step's id, type, name and a one-line summary) and edges (from, to, and the handle for a button or branch). Pass nodeId (a step's id or name) to also get that step's full data in node. Request headers and credentials in step data are masked as [redacted].

Existing fields are unchanged.

update_node merges instead of replacing

update_node used to replace a step's whole data with what the caller sent, so a call carrying only name erased the step's content. It now merges:

  • text, link and keywords are new inputs for the common edits: a message step's reply text, its outgoing link, and the trigger's keyword list.
  • data is merged into the step. Top-level keys are set; content blocks merge by id; the trigger's config merges by key. To remove a content block, pass removeBlockIds. A key you leave out is kept, not deleted.
  • The write fails with state_error (wait_then_retry) instead of overwriting when the flow is being changed at the same time.

Some content is refused with validation_error because only the owner sets it in the flow editor: anything but the name on a payment step, product selections, the post a scheduled-post automation listens on, and removing a button that has a connection. add_node refuses the same values, and add_edge / delete_edge refuse connections out of a payment step.

A restore point before every edit; scheduled-post automations

  • Before update_node, add_node, add_edge, delete_node, delete_edge, revert_flow_to_version, or a generate_flow rebuild changes a flow, the flow as it stood is saved as a revision if no revision already holds it. get_flow_revisions lists it as "Before AI edit" and revert_flow_to_version restores it. If the restore point cannot be saved, the edit is not made (upstream_error, retriable).
  • revert_flow_to_version now works on drafts only, like the other edit tools. For a live flow, duplicate it first.
  • Editing an automation that was set to turn on with its scheduled post turns that setting off, because only a person can approve what goes live. The tool result carries a notice saying so, and the owner gets an in-app notification.
  • duplicate_flow (and POST /api/v1/flows/{id}/duplicate) no longer copies a scheduled-post link. The copy listens on no post until posts are chosen for it.

Scheduled-post automations in the tools

  • get_scheduled_post returns automation when the post has one: flowId, flowName, state (building, failed, off, armed, live, needs_attention), armed, and problem when it needs a fix.
  • New tool detach_post_automation removes the automation from a scheduled post. It needs posts:write and flows:write, and always asks for confirmation. It is being rolled out gradually, so a call may be refused as not enabled.
  • No tool turns an automation on. That is done by the account owner on the post.

What is wrong with a flow; node and trigger catalogs

  • get_flow returns problems: for a draft, every rule that would stop it being turned on (each with code, message and the nodeId when one step is at fault); for a live flow, its current problem. An empty list means none.
  • The sociahive://node-types and sociahive://trigger-types resources are now generated from the product's own lists, so they include every current step and trigger. One correction: a delay step's delay_time is in milliseconds. The resource used to say seconds.

2026-05-20

Outbound webhooks — deferred

The outbound webhooks surface (/api/v1/webhooks/* CRUD and the events listed previously) is gated behind admin access until the production-grade at-least-once delivery pipeline ships. If you were planning an integration that subscribes to SociaHive events, hold off — the public surface is being redesigned and the timeline is open. We'll announce here when it's ready.

In the meantime, the equivalent integration patterns work today via polling the REST API (e.g. GET /api/v1/posts?status=published&since=<timestamp>) or by reading from the MCP list_* tools.

@sociahive/[email protected] on npm

First public release of the official Node / TypeScript SDK.

  • Resource groups: accounts, posts, flows, analytics
  • Typed request and response shapes
  • Ergonomic SociaHiveError with isAuthError, isRateLimited, isNotFound, isStateError helpers
  • Both CJS and ESM bundles; types field correctly ordered for TypeScript conditional resolution
  • Node 18+
npm install @sociahive/sdk

[email protected] on PyPI

First public release of the official Python SDK.

  • Same resource grouping as the Node SDK
  • httpx-based async-capable client
  • SociaHiveError exception class
  • Python 3.9+
pip install sociahive

Docs site

  • New documentation site at docs.sociahive.com — multi-page Fumadocs site replacing the single-file README
  • Three-doors landing: AI assistants / Developers / No-code operators
  • Setup guide with Tabs for each MCP client (Claude Desktop, Claude Code, Shell CLI, Local stdio)
  • Developer quickstart structured as numbered Steps
  • Cookbook with three runnable recipes — auto-reply comments, CSV scheduling, daily performance recap
  • Flow patterns reference (seven named patterns)
  • Error reference and rate limits pages

Public source repositories


How to track future changes

On this page

2026-11-21Every tool from the start in Claude Code2026-11-20More post options on every plan2026-11-18YouTube comment automations from MCP2026-11-17First comments and thread parts on every planTikTok comments, Duet and StitchPlugs on YouTube2026-11-16One Facebook Page connectionAutomation runs counted on every channel2026-11-15Link clicks in automations count in CTRTikTok uploads over 10 MB and inbox captionsInstagram trial reels on every plan2026-11-11Posts only go to your own accountsAccount handles match only where a tool can work2026-11-10@SociaHive in SlackCalendar for named periods2026-11-07Account and permission checksWebhooks2026-11-06Approvals and post comments in the REST APIX Article titles: 100 characters2026-11-05WhatsApp links without REF_YouTube watch time in get_post_analyticsReels watch time in get_post_analyticsPause public replies on a busy post, or switch them off for one postPlatform options in get_scheduled_postTool groups2026-11-04Fewer tools for the same lookupsClearer tracked-link description2026-11-03Your own message in WhatsApp linksLinkedIn document postsEdit a live flow without changing it for people until you publishUpload several local files in one call2026-11-02Turning on a flow and importing a file from a URL ask for approvalResources need read scopes, and unverified apps are labelledOAuth tokens are bound to one resource (RFC 8707)Tool results mark text written by other people2026-10-31YouTube: upload now, and comment replies, on every plan2026-10-30Flow changes wait while a flow is being builtEvery way of creating and scheduling a post applies the same checksPDF uploads, WhatsApp link checks and contact linkingComment reply texts take turns, and per-post caps are checkedgenerate_flow keeps to what you askedConnect, re-point and remove connections by buttonWhatsApp buttons in more than one messageA branch with nothing connected ends thereYouTube title, visibility, audience and category2026-10-29Upload a local or private file from your AI toolInstagram Reels and Stories for every planCap public comment replies per postTurn on a post's automation from your AI appWhatsApp ref linksEditing a live flowDelete and restore for every planInstagram accounts that need reconnecting to postFlows build only what you ask forConnections from buttons and the follow stepBuild state on get_flow, and edits wait for the buildFixes for AI clients2026-10-28Tool descriptions describe, they don't instructClearer tool choice for AI clients (server 1.1.0)2026-10-27Contacts in webhooks and Zapier2026-10-26Plug and posting slot tools renamedPlainer generate_flow descriptiongenerate_flow takes a nameEvery accountId takes a handleAccounts the user names are checked before anything runs2026-10-25Name an account by its handleList the posts of one connected accountAccount errors say what went wrongAI tools can build an "Ask to follow" stepadd_node no longer accepts type: "facebook"Images from ChatGPT go straight into posts2026-10-24Open the chat widget from your own button2026-10-20SDKs, CLI and the n8n node releasedSlack, Tumblr, Dribbble, Twitch and Kick2026-10-19Scheduler features for every plan, and 8 more platformsPinterest boards, and Instagram Stories and Reels2026-10-15Admin preview: scheduler features2026-10-14A caption and media per platform, and a fix for platform_overridesX Articles (rolling out)2026-10-13URL import errors, and post statuses for older iOS buildsUploading media from the CLI and from a URLScheduling and postsMore connected accounts, and a limit on automated accounts2026-10-11MCP tools carry titles and hints2026-10-09Outbound webhooks are available2026-10-06New tool: export_collected_dataget_collected_data hides email- and phone-shaped previews2026-10-05Read-only access on the consent screenToken revocation endpoint2026-10-02generate_flow: iterateOnFlowId no longer rebuilds a flow a person has editedget_flow returns the flow's stepsupdate_node merges instead of replacingA restore point before every edit; scheduled-post automationsScheduled-post automations in the toolsWhat is wrong with a flow; node and trigger catalogs2026-05-20Outbound webhooks — deferred@sociahive/[email protected] on npm[email protected] on PyPIDocs sitePublic source repositoriesHow to track future changes