Skip to content
PolyglotPolyglot

MCP servers

Any server listed under mcpServers in settings.json is connected over stdio at startup. Its tools are exposed to the model as mcp__<server>__<tool> and go through the exact same text-parsed grammar as the built-in tools - an MCP tool is just as usable by a local model with unreliable native tool-calling as by Claude or GPT.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/an/allowed/directory"]
    },
    "github": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "GITHUB_TOKEN", "ghcr.io/github/github-mcp-server"],
      "env": { "GITHUB_TOKEN": "ghp_..." }
    }
  }
}
  • command + args launch the server process; env (optional) is extra environment for it.
  • If a configured server fails to start, Polyglot logs the error and continues without it - startup is never blocked, and the rest of the CLI works normally.
  • MCP tools obey the same permission modes as built-in tools. Polyglot does not otherwise sandbox a third-party MCP server - it runs with whatever access you grant it.

mcpServers is optional; omit it entirely if you don’t need one.