MCP for social media: let an agent finish the job

A social media MCP server gives an AI agent a small set of tools: list channels, validate a post, create it, schedule it, publish it and read the numbers. The agent stops writing captions for a human to copy and starts finishing the job.

MCP is the Model Context Protocol, an open standard between an AI client and a server. The client can be Claude Code, Codex, Cursor or any other MCP-compatible agent. The server describes its tools and their inputs; the agent decides when to call them.

Why most social media MCP servers stop at the draft

Publishing is the hard part. It needs an OAuth token for every platform, a media upload, a validation step for each network and an error path when an API changes. It is much easier to build a server that only reads public data or returns a draft.

That is why lists of social media MCP servers often mix two different things: servers that fetch trends and comments, and servers that can actually publish. Before connecting an agent, check which one you are looking at.

What a publishing server needs

• Official APIs. Shared logins and browser automation are not a stable base and put the account at risk.
• Server-side credentials. The agent should never see a social network password or a raw OAuth token.
• An approval step. The owner reviews the draft before it goes out, with one tap in Telegram or email.
• A live channel list. A server may support ten networks in code and three in production while the rest wait for platform review. Ask for the live list.
• Clear errors. When a platform rejects a post, the agent must receive a reason it can act on, not a silent success.
• Analytics. Publishing without reading views, comments and clicks is half a workflow.

What the Postmately MCP server exposes

The endpoint is https://api.postmately.com/mcp with bearer authentication. One key sees only the channels of its workspace.

Tool group	What the agent can do
Channels	List connected channels, see which ones are live and which are waiting for review
Validation	Check a post against the limits of each platform before creating it
Posts	Create, update, reschedule, delete and read a post
Media	Make a carousel of 1080x1920 slides or a vertical video with a hook and captions
Analytics	Read post metrics and channel insights

Live channels: TikTok, YouTube, Instagram, Facebook, Threads, Telegram, VK, Pinterest, Bluesky and Mastodon. LinkedIn is waiting for platform review, so it is not available for publishing yet.

Setup in five minutes

Create an API key in the dashboard, then add the server to your MCP client. This is the Claude Code command:

claude mcp add --transport http postmately https://api.postmately.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

For a JSON config, the same server looks like this:

{
  "mcpServers": {
    "postmately": {
      "type": "http",
      "url": "https://api.postmately.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Then give the agent the rules of the platforms. The skill file at /skills/postmately/SKILL.md lists the tool order, the limits of each network and the approval rules. Without it the agent knows the tools but not the workflow.

What the agent can do end to end

1. Read the website and build a 30-day plan.
2. Create three carousels and one short video from the owner's photos.
3. Validate every post against the limits of VK, Telegram and TikTok.
4. Send the drafts for approval in Telegram.
5. Publish approved posts through official APIs.
6. Read views, comments and clicks, then adjust the next batch.

Before you connect an agent

• Ask for the live channel list, not the roadmap.
• Confirm that the server holds the tokens, not the prompt.
• Check where approval sits and how to revoke it.
• Test one post on one channel before scheduling a month.
• Make sure the agent receives analytics, otherwise it cannot learn.

Next: read the Claude Code setup guide, see the Telegram scheduler or the AI social media autopilot.

Start free and connect an agent