lm

lm is a local MCP server: Connect Claude to any OpenAI-compatible LLM endpoint and offload routine work to a local model. Install it from npm as @houtini/lm.

Claude CodeCodexAntigravity

commandprompt

Terminal

claude mcp add lm -e HOUTINI_LM_ENDPOINT_URL=... -e HOUTINI_LM_API_KEY=... -e HOUTINI_LM_MODEL=... -e HOUTINI_LM_PROVIDER=... -- npx -y @houtini/lm

Run in your project folder. Add --scope user to use it everywhere.

Add to claude_desktop_config.json or .cursor/mcp.json.

{
  "mcpServers": {
    "lm": {
      "command": "npx",
      "args": [
        "-y",
        "@houtini/lm"
      ],
      "env": {
        "HOUTINI_LM_ENDPOINT_URL": "...",
        "HOUTINI_LM_API_KEY": "...",
        "HOUTINI_LM_MODEL": "...",
        "HOUTINI_LM_PROVIDER": "..."
      }
    }
  }
}

Fill in the environment variables with your own credentials.

• Runtime: Node.js 18+ (runs with npx).
• MCP client: Claude Code, Codex, Antigravity, Claude Desktop, Cursor or any other client that speaks MCP over stdio.
• Environment variables: HOUTINI_LM_ENDPOINT_URL, HOUTINI_LM_API_KEY, HOUTINI_LM_MODEL, HOUTINI_LM_PROVIDER