Skip to main content

Changelog

All notable changes to the PostEverywhere API.

2026-06-14 — Drafts via API & MCP: create-review-schedule for agents​

You can now create a post as a draft through the API and MCP, review it, and schedule it in a separate step. This unlocks the human-in-the-loop approval pattern: an agent drafts posts, a person (or a second agent) reviews them, and only then are they scheduled or published. All changes are additive — existing integrations are unaffected.

Draft creation on POST /v1/posts​

  • POST /v1/posts now accepts draft: true — saves the post as a draft instead of publishing or scheduling it. The response returns status: "draft", the post_id, and a next_steps hint pointing at the schedule endpoint.
  • account_ids is optional for a draft. You can save a draft now and choose its target accounts later (either when editing the draft or in the schedule call). For a non-draft post, account_ids is still required.

New endpoint: POST /v1/posts/{id}/schedule​

  • Turns a draft into a scheduled (or immediately-publishing) post. Body: { "scheduled_for": "<ISO 8601 UTC>" } to schedule, or { "publish_now": true } to publish immediately. Optional account_ids (to set/override the draft's targets) and timezone (display metadata).
  • Only works on drafts. Calling it on a post that is already scheduled, publishing, or published returns 409 — re-scheduling an existing scheduled post is still done with PATCH /v1/posts/{id}.

Richer drafts on GET /v1/posts and GET /v1/posts/{id}​

  • GET /v1/posts?status=draft and GET /v1/posts/{id} now return each draft's target account_ids and platform_content, so an agent (or a reviewer) can inspect exactly what a draft will publish before scheduling it.

MCP server​

  • create_post gains a draft boolean — set draft: true to save a draft instead of publishing/scheduling.
  • New schedule_post tool — params: post_id, plus scheduled_for (ISO 8601 UTC) or publish_now: true, and optional account_ids and timezone. The full agent flow is now create_post(draft: true) → review with list_posts(status: "draft") / get_post → schedule_post.

ChatGPT & OpenAI Codex​

  • The hosted MCP server and the agent docs now call out ChatGPT (via the hosted connector at posteverywhere.ai/agents) and OpenAI Codex alongside Claude Code, Claude Desktop, and Cursor as supported AI clients.

2026-06-11 — Major API upgrade: introspection, webhooks, campaigns, bulk ops​

The largest single API expansion since launch. Eleven new endpoints + a comprehensive enhancement of GET /v1/posts. All changes are additive — existing integrations continue to work unchanged.

New endpoints​

  • GET /v1/me — introspection. Returns your API key context, scopes, organization, quota (accounts/AI credits/storage), and plan limits in one call. Call this first when initializing an integration so you don't have to hardcode organization_id.
  • GET /v1/analytics/summary — aggregate counters for a time window (today / week / month / all / custom). Returns posts-by-status, per-platform breakdown, total views/likes/comments/shares/impressions/clicks, and AI credits used. One call instead of listing every post and counting client-side.
  • Campaigns CRUD (GET/POST /v1/campaigns, GET/PATCH/DELETE /v1/campaigns/:id) — group related posts under named campaigns. Tag posts via campaign_id on POST /v1/posts and filter via ?campaign_id=… on the list endpoint.
  • POST /v1/posts/bulk — create up to 50 posts in one call. Counts as ONE API-rate-limit hit instead of N. Returns per-item success/error so you can handle partial failures (HTTP 207 Multi-Status for mixed outcomes).
  • POST /v1/posts/retry-failed — bulk retry every failed destination matching a filter (account_id, platform, date range, post_ids, max_attempts). Refuses when no filter is provided to avoid "retry the entire failure history" footguns.
  • GET /v1/accounts/:id/health — token expiry, last successful publish, recent failure count, needs_reconnection flag, can_post boolean, and an overall status (healthy/warning/broken). Use this BEFORE publishing to detect a dead token in advance.
  • POST /v1/accounts/:id/reconnect — generate a signed dashboard URL that opens the platform OAuth reconnect flow for an account. 30-minute single-use token.
  • POST /v1/ai/generate-caption — generate 1-5 caption variants from topic + tone + platform. Respects per-platform character limits (X: 280, Bluesky: 300, LinkedIn: 3000, IG: 2200) and hashtag conventions. Costs 1 AI credit per returned caption.
  • Webhooks — POST/GET /v1/webhooks, GET/PATCH/DELETE /v1/webhooks/:id, POST /v1/webhooks/:id/test. Subscribe to event streams instead of polling. 12 event types (post.published, post.failed, post.partially_failed, account.reconnect_needed, media.uploaded, …). HMAC-SHA256 signing, exponential-backoff retries (30s → 2min → 10min → 1h → 6h → 24h, 6 attempts), auto-disable after 20 consecutive failures.

Enhanced GET /v1/posts​

All new query params are optional and additive — the old single-value ?status= and ?platform= keep working.

  • status now accepts comma-separated values, and adds failed, partially_failed, publishing (was: only scheduled / published / draft). Example: ?status=failed,partially_failed.
  • platform now accepts comma-separated values. Example: ?platform=instagram,tiktok.
  • account_id — filter to one social account.
  • campaign_id — filter to a campaign.
  • Date range filters: created_after, created_before, scheduled_after, scheduled_before, published_after, published_before, updated_after (the last one for incremental sync polling).
  • search — full-text search on post content (ILIKE).
  • sort — created_at (default) | scheduled_for | published_at | updated_at.
  • order — asc | desc (default).
  • Response includes pagination.total + pagination.has_more so clients know if more pages exist without fetching the next one.
  • Response includes per-destination metrics (views/likes/comments/shares/impressions/saves/clicks) inline — no more N+1 fetch pattern.
  • Response includes campaign_id on each post for round-trip + filtering.

MCP server expansion​

The Claude Code MCP server now exposes the new endpoints as tools: get_me, get_analytics_summary, list_campaigns, create_campaign, get_campaign, update_campaign, delete_campaign, bulk_create_posts, retry_failed_posts, get_account_health, reconnect_account, generate_caption, list_webhooks, create_webhook, get_webhook, update_webhook, delete_webhook, test_webhook, list_posts_advanced. Existing tools (list_posts, create_post, etc) are unchanged.

SDK​

@posteverywhere/sdk v1.3.0 publishes companion methods: client.getMe(), client.getAnalyticsSummary(), client.listCampaigns(), client.bulkCreatePosts(), client.retryFailedPosts(), client.getAccountHealth(), client.getAccountReconnectUrl(), client.generateCaption(), client.listWebhooks(), client.createWebhook(), client.listPostsAdvanced(). All new methods are TypeScript-typed.

Backwards compatibility​

  • All previously documented params and response fields continue to work and contain the same shape.
  • ?status=scheduled (single value) still works — comma-separated is opt-in.
  • ?platform=instagram (single value) still works.
  • Old pagination: { limit, offset } shape unchanged; new fields (total, has_more) are added.
  • New response fields (campaign_id, metrics, etc) are appended; nothing was removed.
  • New endpoints are net-new — they don't collide with anything in v1.

2026-05-29 — Bluesky support, accurate post status, media auto-finalize​

Bluesky (AT Protocol) is now a supported platform — the 9th. Connect a Bluesky account with a handle + app password, then publish text, images (up to 4), video (MP4), and threads. Links, @mentions, and #hashtags are turned into rich-text facets automatically, and URLs render as link cards. See the Bluesky guide.

  • post_status now reflects real publishing progress. Previously a publish_now post could report scheduled indefinitely because the value was fixed at creation time. The status is now derived from per-destination state and can be scheduled, publishing, published, partially_failed, or failed — consistent across POST /v1/posts, GET /v1/posts/{id}, the list endpoint, and /results. The ?status= list filter now filters on the same real status.
  • Posts auto-finalize uploaded media. If you attach media_ids for media that was uploaded but not yet finalized with POST /v1/media/{id}/complete, POST /v1/posts now completes it for you (verifying the file exists) instead of returning media_not_ready. The explicit /complete call is still supported and recommended for tighter control.
  • Clearer errors. Posting multipart/form-data to /v1/posts now returns guidance on the correct JSON shape and the upload → media_ids flow. The media status response exposes ready and action_required fields so polling clients know a /complete call is required.

2026-04-12 — scheduled_for rename, strict field validation, scope visibility​

scheduled_for is now the canonical request field for scheduling posts. The old name scheduled_at was a misnomer — it read as a past-tense timestamp (like created_at / updated_at) even though it represented a future moment when the post will publish. scheduled_for makes the intent explicit.

  • scheduled_for is the canonical request field name (was scheduled_at). The response also returns the value under scheduled_for, completing the round-trip.
  • scheduled_at is deprecated but still accepted as an alias. Existing integrations built before 2026-04-12 will continue to work unchanged. New integrations should use scheduled_for.
  • Strict 400 validation for unknown media field names. If you send media, media_id, mediaId, or attachments in a request body, the API now returns 400 invalid_field_name with a helpful did you mean media_ids? hint. The only accepted media field on requests is media_ids (an array of UUID strings).
  • OAuth scope visibility. When a user reconnects a social account and the platform drops scopes from the grant, we now record the drop and alert on the mismatch, so silently-broken integrations surface quickly.
  • Body logging for v1 endpoints. The v1 API now logs sanitized request bodies (with API keys and tokens redacted) for debug visibility, making it much easier to trace bad requests when you email developer support.
  • Round-trip principle. Every top-level field name in a request body also appears in the response with the same name and meaning. You can take a post you got from GET /v1/posts/{id} and POST its fields straight back to clone it — no renaming.

v1.0 — March 2026​

  • Initial public API release
  • 18 endpoints across Accounts, Posts, Media, and AI
  • Support for 8 platforms: Instagram, TikTok, YouTube, LinkedIn, X, Facebook, Threads, Pinterest
  • Bulk scheduling (up to 100 posts per request)
  • AI image generation
  • MCP server integration for Claude Code
  • Node.js SDK

Subscribe to API updates by emailing [email protected].

Resources: