add_post_note
ChatGPTAdds a note for the team on a post, up to 5,000 characters: shown with the post in the app, never to the client reviewer and never published. get_approval lists the latest notes. Use it to explain a choice or leave your human a question.
cancel_invitation
ChatGPTCancels an invitation that is still waiting for an answer (list_members invitations): its link stops working at once. Needs the publish scope and a key whose human manages the team, like invite_member. Removing a member or changing someone's role isn't offered to agents (a misled agent could lock people out or hand out write access): the human does it on the Team page. (Not available in an unclaimed sandbox.)
cancel_post
ChatGPTUnschedules a post that has not started publishing (it stays as an approved draft you can schedule again).
complete_media_upload
ChatGPTAfter the PUT to upload_url: checks that the stored file has exactly the declared size and type (otherwise the upload is dropped and you start again with create_media_upload), measures a video (width and height as displayed, duration_ms), and makes the media ready for create_post.media_ids. width, height and duration_ms are fallbacks, used only when a video can't be measured. Calling it again returns the same media.
create_auto_reply
ChatGPTAdds a rule that answers comments or direct messages automatically, on the person's behalf, from now on (never older comments). trigger COMMENT: when someone comments (optionally only with one of keywords, whole words, any case; only on post_ids; only on channel_ids), post one of public_replies (picked at random) under the comment and/or send message (+ link_url as a tracked short link) privately to the commenter, once per comment within 7 days. trigger MESSAGE: when someone sends a direct message (optionally with a keyword), answer with message within 24 hours. once_per_person (default true) answers each person once per rule per 30 days. At most 120 replies per account per hour. This sends messages to real people without a human in the loop: show the person the exact texts and get their OK before creating it enabled, or create it with enabled=false. (Not available in an unclaimed sandbox.)
create_connect_link
ChatGPTA link to send the account owner (your client or your human). They open it, sign in to the network and pick their accounts — no Shager account needed. The accounts then appear in list_channels. Live: YouTube channels, X (Twitter) accounts and LinkedIn personal profiles. Facebook: Awaiting Meta's approval: until then, only people we gave early access can connect. Pages only: Facebook lets no app publish to personal profiles. Posts, photos, video, Reels and Stories, with your own cover image for videos. First comment with a clickable link. Instagram: Awaiting Meta's approval: until then, only people we gave early access can connect. Connected directly with Instagram Login or through a Facebook Page. Your own cover image for Reels. Links in captions and comments aren't clickable on Instagram: the tracked link goes in the first comment as text, and a clickable link can only go in the profile. TikTok: Pending TikTok's app review: until approved, posts can only go to private accounts and are published as private ("only me"). Privacy, interactions and commercial-content disclosure are chosen per post. TikTok takes no cover image and applies a chosen cover moment to public posts only, so Shager can put your cover image at the start of the video. (Not available in an unclaimed sandbox.)
create_group
ChatGPTCreate a group inside one workspace: a named set of its channels, so a post can target all of them at once. channel_ids must belong to this workspace.
create_link
ChatGPTMakes a tracked short link for any web address, for a bio, a story, an email or an ad: each click is counted (list_links) and the visitor lands on the address with UTM tags added. source and medium default to shager and link, campaign to none; words become lower-case-with-dashes. For a link inside a post use cta_url on create_post instead: every account then gets its own link, tagged automatically.
create_media_upload
ChatGPTFor a file on your machine that has no public URL: reserves the space and returns media_id and a signed upload_url. PUT the file's raw bytes there with exactly the returned headers (content-type, content-length, if-none-match: *) within 30 minutes, then call complete_media_upload with the media_id; until then the media can't go on a post. The file goes straight to Shager's storage, not through this API. Types: JPEG, PNG, WebP, GIF, MP4, MOV, WebM and PDF, up to 2 GB; size_bytes must be the file's exact size. For a file that is already online, upload_media_from_url does it in one call.
create_pinterest_board
ChatGPTCreates a public board on a Pinterest account (name up to 50 characters) and returns it ({id, name, privacy}) for per_channel.pinterest_board_id. The board shows on the person's public Pinterest profile at once, so create one only when your human asked for it. At most 10 new boards an hour per account, and per person. (Not available in an unclaimed sandbox.)
create_post
ChatGPTOne post, many accounts. Pick accounts with channel_ids and/or group_ids from one workspace; the post gets one destination per account, each with its own format, text override and network fields (per_channel). Without schedule_at the post is saved as a DRAFT and the response lists every validation issue (issues[]) so you can fix them with update_post; nothing is published. With schedule_at it is validated (all issues at once), checked against the key's caps and the X link allowance, and scheduled — then every account publishes at its own staggered time. Formats: default per network (Facebook/Instagram FEED, REEL with a video; YouTube SHORT; TikTok VIDEO; LinkedIn FEED, VIDEO with a video; X FEED). An unsupported format is an issue, never swapped silently. Links: cta_url becomes a tracked short link per account with UTM tags — in the caption where links are clickable (Facebook, LinkedIn, YouTube) and in the first comment on Instagram. first_comment is a comment on Facebook/Instagram/YouTube and LinkedIn company Pages, a reply on X, and skipped on LinkedIn profiles and TikTok. On X nothing is added automatically and any link you write (text or reply) costs one of the plan's monthly X link posts (whoami shows what is left). TikTok: per_channel.tiktok_privacy is required per account (values from list_channels); comments/duet/stitch default to off; set tiktok_commercial with brand_organic/brand_content for ads. YouTube: youtube_title/privacy/made_for_kids, overridable per channel. Time: schedule_at is ISO 8601 with an offset (2026-10-20T09:00:00+03:00); its exact UTC instant is preserved; the workspace timezone is for display. Staggering: accounts on one network go out one after another with a random stagger_min_seconds–stagger_max_seconds pause (default 30–90 s; Facebook and Instagram share a lane). Sends to more than 10 accounts or within 60 minutes need confirmation_token from the returned plan (you first get needs_confirmation with the plan). Approval: if the workspace has approval_required, create the draft, request_approval, wait for get_approval to show approved, then schedule_post. Vary the caption per account — identical mass posts look like spam. With schedule_at the post goes out publicly on the person's real accounts at that time: only schedule what your human asked to publish.
create_posts
ChatGPTUp to 25 posts of one workspace in one call: a week or a month of content, or one piece per account at each account's best time (get_best_times next_times). Each item in posts is exactly a create_post body without workspace_id (channel_ids/group_ids, caption, media_ids, schedule_at, per_channel, fit_shape, confirmation_token…), and goes through create_post one by one, in order, with all its checks, caps and confirmations. One item's failure doesn't stop the others: results has one entry per item, in order, with ok and either the post or the error (code, message, details). An item that needs confirmation comes back with error.code needs_confirmation and its confirmation_token in error.details: show the plan to the human, then send only those items again with confirmation_token. validate_only: true runs validate_post on every item instead and saves nothing (each result is a validate_post answer with draft_id and confirmation_token). Retrying: give each item an idempotency_key, or one idempotency_key for the call (item i then uses "<key>:<i>"), and a retry replays the items that went through and retries the rest.
create_webhook
ChatGPTInstead of polling get_post, have Shager POST events to your HTTPS endpoint (a CRM, your agent's server): post.destination.published (data: post_id, destination_id, channel_id, network, account, format, remote_url, remote_post_id, published_at, notice), post.destination.failed (… error, unconfirmed), post.completed (post_id, status, counts, destinations[] with state and remote_url), approval.decided (post_id, approval_id, decision, post_status, reviewer.name, comment). Body: { id, type, created_at, data } (data always has workspace_id). The id is the same on every retry: dedupe on it. Order is not guaranteed. Headers: Shager-Event-Id, Shager-Event-Type and Shager-Signature: t=<unix seconds>,v1=<hex HMAC-SHA256 of "<t>.<raw body>" with the secret>. Verify it on the raw body in constant time and refuse timestamps older than 5 minutes. Answer 2xx within 10 s. Anything else (redirects included) is retried: 6 attempts over about a day. An endpoint whose every attempt fails for 3 days is turned off; a successful send_test_webhook turns it back on. Signing secret (whsec_…): the REST API returns it once, in this response. Through an AI assistant it is never returned: the person copies a fresh one in the app (the workspace's Settings → webhooks → "New signing key") and puts it in the receiving system. https only; private, local and credentialed URLs are refused. Up to 10 endpoints per workspace. Publishing never waits for webhooks. (Not available in an unclaimed sandbox.)
create_workspace
ChatGPTCreate a workspace for a new client (agency) or brand. Its accounts are then connected inside it (create_connect_link). Its timezone is how schedule times without an offset are read and how dates are shown.
delete_auto_reply
ChatGPTDeletes a rule; it stops answering at once. Its activity stays in list_auto_reply_activity (kept 90 days).
delete_group
ChatGPTDeletes a group. Its accounts and posts stay; posts just stop remembering the group as a target. Cannot be undone through the API (create_group makes a new one).
delete_post
ChatGPTDeletes a post that is not scheduled and never published (DRAFT, IN_REVIEW, CHANGES_REQUESTED, APPROVED, or a canceled post). A pending approval link stops working. Scheduled posts: cancel_post first. Published posts stay for their analytics. Cannot be undone through the API.
delete_webhook
ChatGPTStops sending events to this endpoint, including retries still pending. Cannot be undone.
get_analytics
ChatGPTHow a workspace's posts did in a period: views, reach, likes, comments, shares, saves, engagement (likes + comments + shares + saves) and engagement rate, clicks on Shager's tracked links, and followers — in total, compared with the previous period of the same length, day by day, by network, by account, and the top posts. Numbers come from each network's own API and refresh every few hours (often in a post's first day, then less). A post counts in the period it was published in, with its latest numbers. Totals add up what the networks report: Facebook views appear once Meta approves read_insights, LinkedIn shares numbers only for company Pages (impressions as views, unique impressions as reach, likes, comments, shares; followers) and none for personal profiles, and a TikTok account counts only after it is reconnected with the video-reading permission and only for its public videos (otherwise by_network.measured=false, coverage.notes) — only their link clicks count. get_post shows one post per account, with null for a number the network doesn't report. Use it to answer 'what worked this week', to pick the best time or format, or to report to a client. Filter with channel_ids or group_ids.
get_approval
ChatGPTThe latest approval request for a post: decision (PENDING, APPROVED, CHANGES_REQUESTED, REJECTED), the reviewer, when it was opened and decided, and the reviewer's comment.
get_best_times
ChatGPTThe best hours of the week to post, for each account of a workspace, with the next times they come round (ready for schedule_at). Each account is timed for its own audience: one country, several, the whole world, or a time zone — from this request (countries, world, timezone), else the account's market, else its groups' markets together (set_audience), else where its clicks come from, else the workspace's zone; audience says which and how each country is weighed. One market is read on its own clock; several markets on UTC, with each hour also given in every market's local time, so a worldwide group gets the hours that reach most of its audience. A worldwide audience with no clicks yet is weighed by an assumed mix (audience.assumed_mix). Every source Shager has comes back, most specific first, and best is the first with enough data: account_posts (the account's own posts of the last year, scored by likes, comments, shares, saves and clicks against its typical post; needs 12 measured posts), audience_clicks (when its audience clicked the posts' short links; 50 clicks, else workspace_clicks), shager_country (when people in each of its countries click Shager short links across all customers: counts per hour only, a country only above 500 clicks from 5 workspaces in 90 days), and study (Buffer's 2026 study of millions of posts per network, read in each market's local time; none for Pinterest, Bluesky, Telegram, Mastodon or Google Business). missing says why a more specific source wasn't used. To post at each account's best time without picking it yourself, give create_post, create_posts or schedule_post schedule_at "best": Shager picks every account's next free best time (one post, a delay per account) and says what it picked in best_time.
get_claim_link
ChatGPTSandbox only. Issues a new claim link for your human (the one from signup stops working) and emails it again to contact_email when there is one. Call it when you lost the signup response.
get_holidays
ChatGPTHolidays of every major tradition (Jewish/Israeli, Muslim, Christian, Orthodox, Druze, Bahá'í, Hindu, Buddhist, Sikh, Jain, Chinese, Persian, international and commercial days) with advice per day: avoid (no promotional posts to that audience), sensitive (respectful only), opportunity (greeting or themed post), normal. Check this before scheduling, and match the traditions to the audience of each account.
get_insights
ChatGPTWhat went with each account's better and worse posts, as findings to act on in the next post: video or photo, text alone, carousels, a question in the first line, a link, short or long text, 4+ hashtags, an emoji up front, weekends, and the format when it used several. Each finding gives the trait, better or worse, lift (its posts' typical result against the others', 2 = twice as good), how many posts are on each side, and advice in plain words. Also trend (the last 5 posts against the ones before: rising, falling or steady) and top_posts. Posts are scored against the account's own typical post by likes, comments, shares, saves and clicks on Shager's short links (views where a network gives nothing else), so findings are the account's, not an average of strangers. Needs 8 measured posts; a finding needs 3 posts on each side and a 25% difference. missing says what's short. These go together with results, they don't prove cause: try a finding and check again. Use it before writing the next posts; get_analytics has the numbers, get_best_times the hours.
get_network_features
ChatGPTThe feature × network table the site shows on /networks, built from the same code: for each kind of account (Facebook Page or profile, Instagram, Threads, YouTube, X, LinkedIn profile or company Page, TikTok, Bluesky, Telegram, Mastodon, Pinterest, Google Business Profile) whether it is open to everyone today, publishes automatically, takes a first comment and a video cover, lets you read, answer, hide and delete comments, runs auto-replies and private messages, reports post numbers, and whether removing it in Shager also cancels access at the network. value: yes, limited (see note), later (built, waiting for the network's approval of Shager), no (the network doesn't allow it). Check it before promising a feature to a person.
get_post
ChatGPTStatus of a post and of every account it goes to: state, dispatch and publish time, live URL or error, and once published its results (views, likes, comments, shares, saves, link clicks; null = not reported by that network) — plus the remaining validation issues while it is still a draft, and next_steps.
get_structure
ChatGPTHow a Shager account is organized: - Organization: the whole account, with one plan and one team. Your key belongs to exactly one organization. - Workspaces: inside the organization. In the app a workspace is called a client when the organization is an agency (one workspace per client business) and a brand or simply "my accounts" when a person or business posts for itself (usually one workspace). Everything below lives inside one workspace and never crosses into another. - Channels: the connected social accounts of that workspace — one Facebook Page, one Instagram account, one YouTube channel, one X account, one LinkedIn profile or company Page, one TikTok account each. A channel belongs to exactly one workspace. - Groups: named sets of that workspace's channels (for example "Hebrew pages" or "North branches"). A channel can be in several groups or in none. Targeting a group posts to all of its ACTIVE channels. - Media, posts, tracked links, webhooks and the client-approval setting also belong to the workspace. - A post belongs to one workspace and targets channel_ids and/or group_ids from that same workspace. Call get_structure (GET /structure) to see the whole tree in one call: every workspace with its channels and groups. When you talk to the person, use their words: "client" (agency) or "your accounts" (one business), and "group". Returns the organization, then every workspace (client or brand) with its connected channels and its groups (each group lists its channel_ids; each channel lists the groups it is in). Use it before posting to pick the right workspace, channel_ids and group_ids.
invite_member
ChatGPTEmails a person an invitation to join this organization's team as an editor or viewer, for every workspace (scope "all") or only some (workspace ids from list_workspaces). The invitation link works for 7 days; inviting an address that already has an open invitation updates and resends it. Sends email to a person, so it needs the publish scope and a key whose human manages the team. Agents can't invite managers: the human does that in the app (Team page). At most 30 invitation emails per organization per day. (Not available in an unclaimed sandbox.)
list_auto_replies
ChatGPTThe workspace's automatic-reply rules, oldest first — the first enabled rule that matches a comment or direct message answers it — each with its last-30-days numbers, plus what every account can do: comments (public replies, a private message to the commenter) and direct messages. Instagram: comments and private replies work; answering direct messages needs a reconnect after Shager added the messages permission. Facebook Pages: comments, public replies and private replies work (comments are read every 5 minutes until Meta delivers them instantly); Messenger messages are answered once Meta delivers them to Shager (after Meta's app review). YouTube and LinkedIn company Pages: public replies to comments only (no private messages; LinkedIn has no messaging API for Pages, and Page comments are checked every 2 hours). X: public replies and DMs, billed to the workspace's monthly X limit. LinkedIn profiles and TikTok: not available.
list_auto_reply_activity
ChatGPTThe latest comments and direct messages that reached the workspace's auto-replies, newest first: who wrote what, which rule answered, the public reply and private message sent, or why nothing was sent (status SENT, SKIPPED, FAILED or PENDING). Comments no rule was meant for are left out. Kept 90 days.
list_channels
ChatGPTThe connected social accounts (channels) of one workspace (client or brand). A channel is one social destination: a Facebook Page, an Instagram professional account, a YouTube channel, an X account, a LinkedIn profile, a LinkedIn company Page (account_kind "company") or a TikTok account. Only status ACTIVE accounts can receive posts; NEEDS_RECONNECT means the human must reconnect it. Each row tells you the network's rules: formats, caption_limit, whether a first_comment exists (per account: LinkedIn company Pages yes, LinkedIn profiles no), whether links in the caption are clickable, requires_media, and paid_links (X: every post comes from the monthly allowance in x_posts_this_month, and a post with a link also from x_link_posts_this_month). TikTok rows carry tiktok.privacy_options — the only values allowed in per_channel.tiktok_privacy — or null with a note when TikTok could not be asked. Live: YouTube channels, X (Twitter) accounts and LinkedIn personal profiles. Facebook: Awaiting Meta's approval: until then, only people we gave early access can connect. Pages only: Facebook lets no app publish to personal profiles. Posts, photos, video, Reels and Stories, with your own cover image for videos. First comment with a clickable link. Instagram: Awaiting Meta's approval: until then, only people we gave early access can connect. Connected directly with Instagram Login or through a Facebook Page. Your own cover image for Reels. Links in captions and comments aren't clickable on Instagram: the tracked link goes in the first comment as text, and a clickable link can only go in the profile. TikTok: Pending TikTok's app review: until approved, posts can only go to private accounts and are published as private ("only me"). Privacy, interactions and commercial-content disclosure are chosen per post. TikTok takes no cover image and applies a chosen cover moment to public posts only, so Shager can put your cover image at the start of the video.
list_connect_links
ChatGPTThe workspace's connect links that still work (not revoked, not expired): who each is for (label), its url to send again, when it expires (14 days after it was made), when it was last used and how many accounts were connected through it. revoke_connect_link stops one at once. (Not available in an unclaimed sandbox.)
list_groups
ChatGPTGroups of one workspace: named sets of its channels (by region, language, branch, campaign…). A channel can be in several groups or none, and a group never includes another workspace's channels. Target a group in create_post with group_ids to post to all its ACTIVE channels.
list_links
ChatGPTThe workspace's tracked short links, newest first (up to 200): each with its short_url, the original URL, its UTM tags, its clicks, when it was made, and the post and account it belongs to (null for links made by hand or with create_link; auto-reply links carry utm_medium auto-reply). Also the totals for the workspace (clicks and unique visitors, bots left out) and clicks by utm_source (unlabeled = links without a source of their own). For clicks per post, account and network in a period, use get_analytics.
list_members
ChatGPTThe people on this organization's team and the invitations still waiting for an answer. role: manager (everything; with scope "all" also the team and plan), editor (writes and schedules posts, answers comments; doesn't connect accounts or change settings) or viewer (read only). scope: "all" (every workspace, including ones added later) or the workspace ids they can open. owner: the team's owner, whose role never changes.
list_notifications
ChatGPTWhat happened lately that the humans were told about (in the app and by email): posts that failed (fully or on some accounts, or only their first comment), accounts that need reconnecting, the X monthly limit reached, auto-reply rules failing repeatedly, accounts a client connected through a connect link, approvals given or changes requested, and team members who joined. Newest first. Each item has a type, an English title and text, the facts in payload, and the app link. Useful to check on posts after scheduling them, or to tell the human what needs them.
list_posts
ChatGPTPosts in a workspace, newest first, with cursor pagination: pass next_cursor as cursor. Filters combine: status, network, channel_id (one account), group_id (posts to any account in that group), query (words in the caption or internal label), and from/to on the scheduled time. Date-only from/to filters use inclusive UTC days, not workspace-local days. Offset timestamps filter exact instants. Posts without a date (unscheduled drafts) only appear without from/to.
list_reviews
ChatGPTThe reviews of one Google Business Profile location (a list_channels account with network GOOGLE_BUSINESS): each review's stars (1–5), text, author, dates and the owner's reply, plus the average rating and the total count. Read live from Google (refresh: true skips Shager's short cache), at most 30 reads per 10 minutes per location and per person. Read only: Google requires the owner's express consent for automated replies, so a person answers reviews on Shager's Comments page, never an agent. reviews is null with a note when Google couldn't be asked.
list_webhooks
ChatGPTThe workspace's webhook endpoints: URL, events, whether it is on (and why it was turned off), the secret's last 4 characters, and the last delivery attempt (time, event type, HTTP status, error).
list_workspaces
ChatGPTA workspace is what the app calls a client (for an agency) or a brand (for a person or business posting for itself; usually just one). Its connected accounts (channels), groups, media, posts and tracked links live in it and never cross into another workspace. approval_required means a client reviewer must approve each post before it can be scheduled (request_approval). For every workspace with its channels and groups in one call, use get_structure.
move_channel
ChatGPTMoves a connected account, with its connection, to another workspace (client or brand) of this organization: nobody signs in again. It leaves the old workspace's groups. Posts already made for it stay in the old workspace (published ones keep their numbers there); posts there that haven't gone out won't go to it any more. The answer gives its channel_id in the new workspace (a new id when the old workspace keeps its history). It changes which client an account belongs to: confirm with your human first. (Not available in an unclaimed sandbox.)
request_approval
ChatGPTThe approval flow: a client reviewer (not a Shager user) gets a link, opens it on their phone without logging in, sees the post per account and approves, asks for changes (with a comment) or rejects. The post moves to IN_REVIEW; get_approval shows the decision; an APPROVED post can be scheduled as long as its content is unchanged. Returns the link to send (and a WhatsApp link when the reviewer has a phone); send_email emails it. Needed when the workspace has approval_required; useful anyway when a human should look first. (Not available in an unclaimed sandbox.)
resend_invitation
ChatGPTEmails a waiting viewer invitation again, with 7 more days to accept. Invitations for editors or managers are resent by a person in the app, so an agent can't revive write access someone let lapse. At most once a minute per invitation, and within the 30 invitation emails per organization per day. Sends email to a person: publish scope and a key whose human manages the team. (Not available in an unclaimed sandbox.)
retry_failed_accounts
ChatGPTSends a FAILED or PARTIALLY_PUBLISHED post again, right now, to the accounts it failed on; accounts it already reached are left alone. Fix the cause first (get_post shows each account's error; often the human must reconnect the account): disconnected or paused accounts stay failed, and an account whose outcome is unknown (the network may have published it) is never retried, so nothing is posted twice. The content is unchanged, so an approval still holds; the plan's monthly limits apply as for scheduling. It goes out publicly at once: retry only when your human wants it. (Not available in an unclaimed sandbox.)
revoke_approval
ChatGPTRevokes the post's pending approval link (the reviewer can't open it any more) and moves an IN_REVIEW post back to DRAFT. A decision already made stays as it is. Send a new link with request_approval.
revoke_connect_link
ChatGPTStops a connect link at once: nobody can connect accounts through it any more. Accounts already connected through it stay connected.
schedule_post
ChatGPTGive an existing draft (or approved post) a time. The post is validated for every account (all issues at once), checked against the key's caps and the X link allowance, and — when the workspace requires approval — must be APPROVED with unchanged content. schedule_at is ISO 8601 with an offset; the workspace timezone is used for display. Each account then publishes at its own staggered time (stagger_min/max_seconds on the post). More than 10 accounts or a start within 60 minutes needs confirmation_token from the returned plan: without it you get needs_confirmation and details.plan (accounts by network, first and estimated last dispatch, X link-post cost, TikTok notes) to show your human. The post then goes out publicly on the person's real accounts: only schedule when your human asked to publish it. Scheduling needs the publish scope and a claimed sandbox. To move a scheduled post, call schedule_post again with the new time; cancel_post stops it.
send_test_webhook
ChatGPTSends a signed webhook.test event to the endpoint now and returns what happened (delivered, HTTP status, error, time taken). Tried once, no retries. A turned-off endpoint that answers 2xx is turned back on. (Not available in an unclaimed sandbox.)
set_channel_status
ChatGPTPauses an account (PAUSED: its connection is kept, nothing new can be scheduled to it, and a post due while it is paused fails for that account, with the reason in get_post) or turns it back on (ACTIVE). Resuming needs a working connection (a NEEDS_RECONNECT account is reconnected by the human, e.g. with create_connect_link) and room in the plan's account allowance. Disconnecting an account isn't offered to agents: it deletes Shager's keys for the account (and, on some networks, cancels access there; see get_network_features) and only the account's owner can connect it again, so the human does it on the Accounts page.
set_client_report
ChatGPTWho gets the monthly report by email (up to 10 addresses; [] stops the emails), its language, and/or a new link. Each recipient gets the link for the month that just ended on the 1st, from 08:00 in the workspace's time zone, with a link to stop getting it. regenerate: true makes a new link and the old one stops working at once: use it when the link reached someone it shouldn't. Shager then emails these people every month on its own, so it needs the publish scope and a claimed organization: only add addresses your human asked for (the client's own, or theirs). (Not available in an unclaimed sandbox.)
set_x_budget
ChatGPTX bills Shager for every read, reply and DM. This sets the most, in US dollars, that this workspace may use on X per calendar month (UTC) for comments and auto-replies; 0 turns X comments and auto-replies off. Default 5. Confirm the amount with your human. (Not available in an unclaimed sandbox.)
update_auto_reply
ChatGPTChanges a rule: send only the fields to change (enabled=false pauses it). The same rules as create_auto_reply apply. (Not available in an unclaimed sandbox.)
update_group
ChatGPTRenames a group and/or replaces its accounts (channel_ids becomes the whole list; [] empties it). channel_ids must belong to this workspace. Posts already made keep the accounts they had. Its market is set with set_audience.
update_post
ChatGPTChange any field of a post that is not scheduled or published (DRAFT, IN_REVIEW, CHANGES_REQUESTED, APPROVED). Send only the fields to change; channel_ids/group_ids replace the account list, per_channel entries are merged field by field into each named account's overrides, media_ids replace the media. Use null to clear cta_url, cta_text, first_comment, label, campaign, cover_media_id, tiktok_cover_ms, cover_first_frame, or a per_channel caption/first_comment override; an empty first_comment suppresses inheritance for that account. Changing content after a client approval sends the post back to DRAFT (approve again). A SCHEDULED post must be cancel_post-ed first. The response lists the remaining issues.
update_workspace
ChatGPTRenames a workspace, changes its time zone (how dates are shown and how times without an offset are read; scheduled posts keep their exact instant), and/or its default UTM tags for tracked links (utm_source, utm_medium, utm_campaign; an empty string goes back to Shager's automatic value). Send only what changes. approval_required can't be changed here on purpose: it is the client's safeguard, so a manager turns it on or off in the workspace's Settings in the app.
upload_media_from_url
ChatGPTDownloads a public image (JPEG, PNG, WebP, GIF), video (MP4, MOV, WebM) or PDF into the workspace and returns a media id to put in create_post.media_ids. Media is per workspace and reusable across posts. Shager measures every video itself: width and height as displayed (a rotated phone video counts as portrait) and duration_ms come back in the response, and validate_post checks Reels, Shorts, Stories and TikTok length and shape rules with them (Shorts: vertical or square, up to 3 minutes; Instagram Reels 3 s–15 min; Stories up to 60 s). width, height and duration_ms are optional and only used when a file can't be measured; measured values win. A video that comes back with null width, height or duration_ms couldn't be measured: pass them and upload again before scheduling it. Private, local and credentialed URLs are refused. Size limit: 2 GB.
validate_post
ChatGPTChecks a post against every account's network rules without saving anything and returns ALL issues at once. It approves nothing by itself: read next_step, which says in plain words what to do (fix issues, ask the human to confirm and pass confirmation_token, or create_post) (caption length, format support, media requirements, video length and shape, YouTube title, TikTok privacy and commercial fields, disconnected accounts). With schedule_at it also reports the X link allowance, the key's caps and whether the send would need confirmation_token from the returned plan, plus the dispatch plan. Run it before create_post when in doubt; it needs only the read scope and works in a sandbox.
whoami
ChatGPTStart here. The organization this key belongs to, its plan and usage (accounts and posts this month, X posts and X link posts left), whether the key may publish (scope, claimed sandbox and a plan in force; why_not says what is missing, and without publishing only drafts work), the key's caps, and — in a sandbox — the claim state. Use can_publish to decide between drafting and scheduling.
cancel_post
Claudecreate_auto_reply
Claudecreate_connect_link
Claudecreate_group
Claudecreate_post
Claudecreate_webhook
Claudecreate_workspace
Claudedelete_auto_reply
Claudedelete_post
Claudedelete_webhook
Claudeget_analytics
Claudeget_approval
Claudeget_claim_link
Claudeget_holidays
Claudeget_post
Claudeget_structure
Claudeinvite_member
Claudelist_auto_replies
Claudelist_auto_reply_activity
Claudelist_channels
Claudelist_groups
Claudelist_members
Claudelist_notifications
Claudelist_posts
Claudelist_webhooks
Claudelist_workspaces
Clauderequest_approval
Claudeschedule_post
Claudesend_test_webhook
Claudeset_x_budget
Claudeupdate_auto_reply
Claudeupdate_post
Claudeupload_media_from_url
Claudevalidate_post
Claudewhoami
Claude