@gtm-api/linkedin-mcp

Launcher for gtm-api: LinkedIn MCP Server, the managed LinkedIn MCP server for AI agents. It bridges a stdio MCP client (Claude Desktop, Cursor, anything that spawns a command) to the hosted streamable-http endpoint at https://mcp.gtm-api.com/mcp via mcp-remote. No server logic runs locally; every tool executes on the hosted server, where account safety is enforced: warm-up, server-side daily limits, preview-then-confirm on outward actions. 20,000+ LinkedIn accounts run on that stack, and under 1% of them have ever been restricted. The method is written up at gtm-api.com/safe-linkedin-automation.

The server exposes three MCP tools (list_toolsets, get_toolset_tools, call_tool) that give an agent progressive access to 160+ typed LinkedIn actions: messaging, connection requests, content, enrichment, search, account health and infrastructure.

Setup

Add the server to your MCP client. There is no key to paste: on first run a browser opens, you sign in and approve access, and the tokens are cached under ~/.mcp-auth.

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "gtm-api": {
      "command": "npx",
      "args": ["-y", "@gtm-api/linkedin-mcp"]
    }
  }
}

Claude Code:

claude mcp add gtm-api -- npx -y @gtm-api/linkedin-mcp

Sign up at app.gtm-api.com (forever free plan, no card) and connect a LinkedIn account you own, either before or during the consent step.

If your client speaks streamable-http itself, skip this launcher and point it straight at https://mcp.gtm-api.com/mcp. It will run the same OAuth flow. The launcher is for clients that can only spawn a command.

Authentication

Two ways in; pick one.

OAuth (default). With no environment set, your client discovers the flow on its own: the first call returns a WWW-Authenticate challenge naming the server's metadata document, mcp-remote registers itself with the authorization server (dynamic client registration, so nothing has to be pre-registered with us) and opens the consent page in a browser. Tokens are cached under ~/.mcp-auth. Best for anything with a human at the keyboard.

API key. Set GTM_API_KEY to a key from the app (Settings, API keys) and the launcher sends it as the bearer on every call: no browser, no consent screen. Best for headless machines, CI and servers where a browser round-trip is not an option. A rejected key fails fast at startup with the reason instead of looping.

{
  "mcpServers": {
    "gtm-api": {
      "command": "npx",
      "args": ["-y", "@gtm-api/linkedin-mcp"],
      "env": { "GTM_API_KEY": "gtm_live_..." }
    }
  }
}

Version notes: launchers up to 1.1.0 required GTM_API_KEY but the endpoint rejected keys back then, so they failed on the first call; 1.2.x ignored the variable and did OAuth only. Key auth works from 1.3.0 on, and requires nothing else in the config.

Environment variables

| Variable | Required | Meaning | |---|---|---| | GTM_MCP_URL | no | Endpoint override, defaults to https://mcp.gtm-api.com/mcp | | GTM_API_KEY | no | Authenticate with this gtm_live_... key instead of OAuth |

Extra CLI arguments pass through to mcp-remote, so its flags are available: --debug to write detailed logs to ~/.mcp-auth, --host to change the callback hostname, --allow-http for a plain-http endpoint override.

Links

• What the server does, tool list, safety model: github.com/gtm-api/linkedin-mcp
• Connecting any client, step by step: docs.gtm-api.com/mcp/connect
• How a LinkedIn MCP server works: gtm-api.com/linkedin-mcp-server
• Docker variant of this launcher: hub.docker.com/r/gtmapi/linkedin-mcp