Best for
- Pi coding-agent users who want lower MCP context overhead
- Developers who need access to multiple MCP ecosystem services
- Users who want lazy connections, metadata caching, or optional direct tool registration
This project is a Pi-focused MCP adapter extension rather than a standalone domain-specific MCP server. Its main value is reducing context usage through one proxy tool, lazy startup, and metadata caching; actual capabilities, permissions, and risks depend on the MCP servers configured by the user.
Pi MCP Adapter is a Pi coding-agent extension that exposes configured MCP services through one proxy tool for on-demand discovery and invocation. It starts servers lazily by default, caches tool metadata, and supports stdio, Streamable HTTP, and SSE fallback to reduce context-window usage.
Run pi install npm:pi-mcp-adapter, then restart Pi. Put project configuration in .mcp.json or shared user configuration in ~/.config/mcp/mcp.json. You can also run pi-mcp-adapter init to scan host-specific configurations.
{"mcpServers":{"chrome-devtools":{"command":"npx","args":["-y","chrome-devtools-mcp@latest"]}}}