MCP Server Setup

PostFuze ships a Model Context Protocol server so AI agents — Claude Code, Claude Desktop, Cursor, Windsurf — can connect social accounts, draft and schedule posts, upload media and read analytics on your behalf. The server is a thin authenticated dispatch layer over the same REST API, so every tool maps onto an endpoint you already know.

Connection details

Authentication uses the same API key as the REST API — generate one in the dashboard and pass it as a bearer token. The key scopes every request to your organization; the agent can only see and act on accounts your key can reach. See Authentication for how to create and rotate keys.

Claude Code

The fastest path. Claude Code speaks Streamable HTTP natively, so you register the server in one command — no proxy required. Replace sk_live_… with your key:

claude mcp add -t http -H "Authorization: Bearer sk_live_…" postfuze https://api.postfuze.com/api/mcp

Verify the connection and list the tools that were registered:

claude mcp list
# postfuze  https://api.postfuze.com/api/mcp  (http)  ✓ connected — 32 tools

From there, just ask in natural language: “Draft a post to my X and LinkedIn accounts and schedule it for 9am tomorrow.” Claude will call list_social_accounts then create_post for you.

Claude Desktop, Cursor & Windsurf

These clients connect to local MCP servers over stdio, so a remote HTTP server is bridged with mcp-remote, run on demand via npx. The config is identical across all three — only the file path differs.

Add a postfuze entry under mcpServers. The --header flag carries your bearer token to the remote server:

{
  "mcpServers": {
    "postfuze": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.postfuze.com/api/mcp",
        "--header",
        "Authorization: Bearer sk_live_…"
      ]
    }
  }
}

Save the file and fully restart the client. PostFuze’s tools then appear in the tool picker. If your shell strips the space in the header, some clients accept the colon-delimited form --header "Authorization:Bearer sk_live_…" instead.

Inspect & debug

Use the official MCP Inspector to browse the tool catalog, view input schemas and fire test calls before wiring the server into an agent. It launches a local web UI:

npx @modelcontextprotocol/inspector@latest

In the Inspector, choose the Streamable HTTP transport, set the URL to https://api.postfuze.com/api/mcp, and add an Authorization header with the value Bearer sk_live_…. Hit Connect, then open the Tools tab to list all 32 tools and the Resources tab to read PostFuze://social-accounts and PostFuze://supported-networks.

Limits & behavior

The MCP server enforces the same constraints as the REST API:

• scheduled_at may be at most 30 days in the future.
• Uploaded media is retained for 60 days after upload.
• List endpoints return 25 results per page by default (max 100). list_postsuses keyset pagination — page by passing the previous response’s next_cursor back as cursor; list_media uses limit + offset. See Pagination & filtering.
• Publishing requires an active subscription; connecting accounts and generating keys are free. Errors are surfaced to the agent as the tool result text.
• Video is a chain — upload_media → confirm_media_upload → wait for ready → create_post — and some platforms need an extra config field at publish (e.g. a Pinterest board). See publishing a video with an agent and per-platform configuration.

Machine-readable docs index

For agents that read documentation directly, PostFuze publishes a plain-text index of every docs page — title, URL and a one-line description — at https://postfuze.com/llms.txt (the llms.txt convention). It is generated from the live docs navigation, so it always lists the current set of pages. Point an agent at it to let it discover the right reference page before calling a tool.

Next steps

• Tool reference — every tool, its parameters and the two resources.
• Posts API — the underlying contract for create_post.
• /llms.txt — a machine-readable index of every docs page for agents.