Connection to a Model Context Protocol (MCP) server.

One Connection process is started per MCP server per request. It fetches the server's tool list on startup and proxies tool calls on behalf of the agentic loop.

Usage

Per-request MCP servers are specified in the API request body:

{
  "model": "gpt-4o",
  "tools": [{"type": "mcp", "server_url": "https://my-mcp.example.com"}],
  "input": [{"role": "user", "content": "Search for recent news"}]
}

Pre-configured servers available to all requests can be set in config:

config :open_responses, :mcp_servers, [
  %{
    name: "docs",
    url: "https://docs-mcp.example.com",
    headers: [{"authorization", "Bearer #{System.get_env("MCP_TOKEN")}"}]
  }
]

See the MCP Integration guide for the full protocol and authentication details.