“How did Friday’s launch post perform?”
Read post impressions, reactions and comments where the network supports them. A fresh reading costs 1 credit; stored results are free.
AI social media agent · MCP + skills
Your AI agent can publish. And measure what works.
Connect Claude, ChatGPT, Cursor or any MCP client to PostMCP once. From then on it drafts, schedules and publishes to seven networks, then reads post analytics. Review drafts and approve publishing in your assistant. Metrics vary by network.
- typed tools
- 17 typed tools
- networks
- 7 networks
- open-source server
- MIT open-source server
ClaudeYou
Runs inside
- Claude
ChatGPT
Claude Code
Cursor
Codex
Gemini CLI
Copilot
Windsurf
Zed
OpenClaw
Hermes
Antigravity
Cline
Goose
OpenCode
VS Code
Connect
Pick your assistant. Three steps, no code.
Every client connects to the same hosted server and sees the same tools. Only the place you paste the connection changes.
Every snippet below has a placeholder where your key goes. Signed in, the dashboard shows the same snippets filled in. Get your key
Claude
Chat app
Add PostMCP as a custom connector on claude.ai. Claude discovers the tools on its own.
- 1
Copy your server URL
The key in the URL is the whole login. Sign in and copy the filled-in URL from Dashboard → AI integration, or paste your key into the placeholder.
MCP server URLURL https://mcp.postmcpai.com/mcp?apikey=YOUR_API_KEY - 2
Add a custom connector
In Claude go to Settings → Connectors → Add custom connector, paste the URL and save.
Open Claude connectors - 3
Ask it something
Start a new chat. Claude will call get_connected_accounts before it drafts.
Try sayingCheck my connected accounts on PostMCP and draft a tweet announcing our new release.
What it can do
Full tool referenceYou talk. It calls the right tools.
“Announce v2.4 on every channel I have connected.”
One post per profile, each with copy fitted to that network’s limit. Nothing publishes until you say so.
“Make a launch image and schedule this for Monday morning.”
The image comes back as a hosted URL, attached as mediaUrl, and the schedule resolves in your timezone.
“Why did this morning’s post fail on X? Fix it.”
It reads the per-profile error, checks the token, and retries only the profile that failed.
Read · safe to run unattended
Write · your client asks first
Guardrails
An agent that can post, not one that posts on its own.
- Drafts wait for you
- Tools that publish are separate from tools that read, so your client can ask before each one. Drafts sit in the queue until you approve them, in chat or in the dashboard.
- The model never sees a credential
- Your social logins stay as encrypted OAuth tokens in PostMCP. The API key lives in your client config. The model gets tool names, arguments and results, nothing else.
- One key, revoked in one click
- Every connection uses the key from your dashboard. Regenerate it and every client loses access at once. Nothing to hunt down per app.
Any client that speaks the Model Context Protocol or loads agent skills: Claude, ChatGPT, Claude Code, Claude Desktop, Cursor, Codex, Gemini CLI, Windsurf, Zed, Cline, Goose, OpenClaw, Hermes and your own agents built on the Anthropic or OpenAI SDKs. They all connect to the same hosted server and see the same 17 tools to publish, schedule and read post analytics. Metrics depend on the network and account permissions.
Only if you tell it to. create_post drafts or schedules by default; publishImmediately has to be set explicitly, and most clients prompt for approval before any mutating tool. Every draft is visible in the dashboard queue where you can edit, reschedule or delete it.
X, LinkedIn, Facebook, Instagram, Threads, Bluesky and YouTube Shorts. The agent targets specific connected profiles by id, so a workspace with two LinkedIn pages can post to one and not the other.
No. The key sits in the connection URL or your client config and authenticates calls to PostMCP. The model only ever sees tool names, the arguments it chose and the results. Social OAuth tokens never leave the backend.
Yes. @postmcpai/server is MIT licensed on npm. Run it over stdio for local clients, or set a PORT to serve it over streamable HTTP from your own infrastructure.
Your assistant is one URL away from your queue.
Create an account, connect your networks, copy the server URL. The first post can be scheduled from a chat window five minutes from now.