XposterAI connector brief (for an AI agent setting up a custom integration)

Server: https://xposterai.com/mcp
Protocol: Model Context Protocol, streamable HTTP, remote. POST only, stateless JSON
  responses (a GET returns 405 by design; there is no SSE stream to open).
Auth: one request header ->  Authorization: Bearer <API key>
  The user copies the key from https://xposterai.com/Account#mcpSection
  Keep it in your secure credential store. Never echo it into chat or logs.
  A wrong or missing key returns HTTP 401. Regenerating the key on that page revokes it.
Also available: OAuth 2.1 with PKCE and dynamic client registration (metadata under
  /.well-known/). Use the API key unless the user asks for OAuth.

What it does: XposterAI publishes and schedules posts to the user's own connected social networks (X, Bluesky, Mastodon, LinkedIn, Pinterest, TikTok) and writes in their trained voice. Rules: call accounts.list first; ALWAYS post.preview before post.publish or post.schedule and tell the user the credit cost; never invent TikTok options — ask; generate a fresh idempotency_key per publish; at most 2 posts per day on X; never put links in X post text unless the user insists (13 credits and downranked). Replies to other people's posts are drafted with voice.write kind="reply" but posted by the user on X, not through this server.

Tools:
- accounts.list: List the social networks connected to this XposterAI account, with per-platform limits. Free. Call this first; "needs_reauth" means the user must reconnect on the website.
- credits.get: Current AI credits and post credits. Free.
- voice.write: Write a post or a reply in the user's trained voice (their persona). Costs 1 AI credit. For kind="reply" pass the tweet being replied to as tweet_text. Returns the text plus any signal-guard warnings (links, bait, generic openers) you should mention.
- post.preview: Validate a post before publishing: per-platform character fit, exact post-credit cost (an X post containing a link costs 13 credits instead of 1), signal-guard warnings, and whether TikTok options are complete. Free. ALWAYS call this before post.publish or post.schedule and tell the user the cost.
- post.publish: Publish now to the given platforms. Spends post credits (see post.preview). idempotency_key must be a fresh random string per user intent — retries with the same key return the original result instead of posting twice. TikTok returns pending=true; check post.status a minute later.
- post.schedule: Queue a post for later. scheduled_for is ISO-8601 WITH timezone offset. Credits are spent when it sends. Refuses a 3rd X post on the same day (author-diversity rule) and TikTok posts without options.
- post.queue: List queued (pending) and draft posts, soonest first. Free.
- post.cancel: Cancel a queued post (drafts are deleted). Free.
- post.status: Outcome of a scheduled post or a single publish record, with URLs. Free.
- radar.find: Trend Radar: find rising posts in the user's lane worth replying to right now (reply early = the small account's best distribution). 3 AI credits, first 5 finds free, results cached 15 minutes (a repeat call inside that window is free), 10 finds/day. Empty results are refunded. Pair with voice.write kind="reply".

Threads: post.preview and post.schedule accept "thread": ["tweet 1", "tweet 2", ...]
  (2-25 items) instead of "text". It is queued as ONE item; every tweet is billed as a
  post. Only networks whose accounts.list entry has supports_threads=true.
Times: scheduled_for must be ISO-8601 WITH a timezone offset, e.g. 2026-10-01T09:00:00-07:00.
Errors: tool failures come back as a normal result with isError=true and a plain-English
  message (for example "Too long for X / Twitter (300/280)."). Show it to the user.
Costs: accounts.list, credits.get, post.preview, post.queue, post.status, post.cancel are
  free. Spending tools say so in their description. Credits belong to the user.

A good first test: call accounts.list, then credits.get, and show both results.
Do not publish or schedule during setup.

Human-readable page: https://xposterai.com/muse   Full tool docs: https://xposterai.com/docs/mcp
Support: support@xposterai.com