Model Context Protocol

Post everywhere from your AI assistant

Connect Claude, ChatGPT or Meta Muse. It drafts in your voice, shows you the cost, and schedules posts, threads, pictures and video to your networks. You approve; it does the clicking.

Your assistant

Posteora

Your networks

Start free Serverhttps://posteora.com/mcp

Posteora ships a Model Context Protocol (MCP) server. Connect it to Claude, Claude Code or ChatGPT and your assistant can publish or schedule to every social network you've connected — X, Bluesky, Mastodon, LinkedIn, Pinterest and TikTok — write in your trained voice, and find rising posts in your niche to reply to.

You draft in the chat as you already do. The assistant finishes with “posted to X, Bluesky and LinkedIn — TikTok scheduled for 9am” instead of “now copy this into your scheduler.”

Endpoint

https://posteora.com/mcp

Streamable HTTP (stateless JSON). Two ways to authenticate:

• Sign in with Posteora (OAuth) — clients that support MCP authorization (claude.ai connectors, ChatGPT, Claude Desktop, Claude Code) discover it automatically: add the URL, click Connect, approve on the consent screen. Tokens are scoped, expire, and can be revoked any time from Dashboard → AI Assistants → Connected via MCP, which also lists assistants using your API key.
• API key — for clients that only take a static header. Find it under Account. Regenerating it signs every assistant out.

Signing in works for every account. The API key route is refused for admin accounts.

Set up

claude.ai / ChatGPT / any OAuth-capable client

Add a custom connector (MCP server) with the URL https://posteora.com/mcp and leave the credentials blank — you'll be sent to Posteora to sign in and approve.

Claude Code (one command, no API key)

claude mcp add --transport http xposterai https://posteora.com/mcp

Then type /mcp inside Claude Code, pick xposterai and choose Authenticate. Your browser opens the Posteora consent screen; sign in, click Allow, and the tools appear. Approve within a few minutes of starting — the sign-in link expires, and /mcp gives you a fresh one if it does.

Claude Desktop / Claude Code config file (API key)

{
  "mcpServers": {
    "xposterai": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://posteora.com/mcp",
               "--header", "Authorization: Bearer YOUR_API_KEY"]
    }
  }
}

Pictures and video (every assistant)

Assistants fail image and video posts when they insist on a public link. Paste these rules once so yours uploads files straight to Posteora (POST /mcp/media) and passes the returned media_id as media_ids:

Posteora media rules. When I attach an image or video that should go out with a post:
- Upload the file yourself, straight to Posteora: HTTP POST https://posteora.com/mcp/media with the same Authorization header you use for the Posteora MCP server (API key or the sign-in token) and the file as multipart form field "file". It returns a media_id. For images you can call the media.upload tool instead.
- Pass that media_id in media_ids when you call post.preview, then post.schedule or post.publish. It stays valid for as long as the post is queued.
- Never ask me for a public link, and never put my files on a third-party host (filebin, imgur and the like). Use media_url only when I give you a link that is already public.
- Limits: JPEG, PNG, WebP or GIF up to 5 MB; MP4 or MOV up to 100 MB (X takes video up to 140 seconds).
Remember these rules for every future Posteora post.

Meta Muse

Muse builds its own connector from one pasted message and your API key. Step-by-step for Muse →

ChatGPT

Add https://posteora.com/mcp as a connector in developer mode with the Bearer header. A directory listing with one-click sign-in is planned.

What to say once it's connected

You need an Posteora account with at least one network connected. Then talk to your assistant normally:

• “What accounts do I have connected?”
• “Draft a post about my launch in my voice.”
• “Schedule that for Friday 9am on X and Bluesky.” It shows you the credit cost before it queues anything.
• “What's in my queue?” / “Cancel that last one.”
• “Write a 10-tweet thread about my launch and schedule it for Thursday 9am.”
• “Post this screenshot to X tomorrow at 9am with a one-line caption.” (attach the file in the chat)
• “Find rising posts in my niche I should reply to.”
• “Schedule that for 9am, then a reply under it at 9:05 with the link.” (reply_to: replies under your own X posts only)

Behind a request like the third one the assistant makes four calls: list accounts, write the draft, preview the fit and cost, then schedule. You only see the question and the answer.

Tools

Same rules as the web

• Every tool calls the same code the Scheduler Studio composer uses. Nothing is cheaper or looser through MCP.
• An X post containing a link costs 13 credits and is downranked — the assistant is told to warn you.
• TikTok posts need your explicit choices (who can view, comments, disclosure). The assistant must ask; it can't guess.
• The assistant may schedule at most two X posts per day. Replies to other people's posts are drafted for you to post on X yourself.
• Publishing requires an idempotency key, so a retrying assistant never double-posts.

Privacy

The server only acts on your own account, with your key. It never reads your assistant's conversation — it sees only the arguments of each tool call. Media passed by URL is fetched once, checked, and stored like an upload. Anything it schedules or publishes appears in your Activity like any other post, and every call is kept in an audit log. Revoke access any time by regenerating your API key.

Connect your networks →

Claude, ChatGPT, Meta and the network logos are trademarks of their respective owners, shown only to indicate compatibility.