MCP servers

Extend the agent with any Model Context Protocol server.

Ask AI about this page

ChatGPT Claude Perplexity Gemini View as Markdown

LocalPilot includes a stdio MCP client. Every tool a server exposes becomes available to the agent as mcp_<server>_<tool>.

Configuration

The Settings → MCP servers page (gear icon in the chat toolbar) lets you add, edit, disable and remove servers and shows each one’s live state — connected with its tool count, disabled, or failed to start with the exact error:

MCP servers page inside VS Code — a connected server with 1 tool, a disabled one, and one that failed to start with the spawn error shown

The same configuration lives under localpilot.mcpServers:

"localpilot.mcpServers": {
  "filesystem": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/dir"]
  },
  "github": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-github"],
    "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_…" }
  },
  "postgres": {
    "command": "uvx",
    "args": ["mcp-server-postgres", "postgresql://localhost/mydb"],
    "enabled": false
  }
}

Field	Purpose
`command`, `args`	Process to launch (stdio transport)
`env`	Extra environment variables (merged with your shell’s)
`cwd`	Working directory (defaults to the workspace root)
`enabled`	Set `false` to keep the config but not start the server

Servers start on activation and restart when settings change. Use LocalPilot: Reload Skills, Subagents & MCP Servers to restart manually; the notification shows how many MCP tools were registered.

Approvals

MCP tool calls follow the localpilot.approvals.network policy (default always). Set it to ask to confirm every call.

Troubleshooting

Open LocalPilot: Show Logs — server stderr is logged under [mcp:<name>], along with initialisation errors and the tool count.