Skip to content

MCP integration

The Model Context Protocol (MCP) is a standard for exposing external tools and data to AI agents. Pi doesn’t speak MCP natively, but extensions add an MCP client so you can plug in the large existing ecosystem of MCP servers (GitHub, databases, browsers, search, and more).

A focused MCP adapter that lets Pi connect to MCP servers and surface their tools to the model.

Terminal window
pi install npm:pi-mcp-adapter

Benefits

  • Purpose-built for MCP — does one thing well.
  • Very popular and actively maintained.
  • Unlocks the entire MCP server ecosystem.

Drawbacks

  • You still configure and run the underlying MCP servers yourself.
  • Each MCP server adds tools (and tokens) to the context.

An MCP plugin focused on context efficiency — it advertises saving a large share of your context window via sandboxed code execution, an FTS5 knowledge base, and intent-driven search. Cross-agent: works with Claude Code, Gemini CLI, Copilot, OpenCode, and Codex too.

Terminal window
pi install npm:context-mode

Benefits

  • Optimizes context usage, not just connectivity.
  • Portable across multiple agents/tools.
  • Built-in searchable knowledge base.

Drawbacks

  • More than a thin adapter — it imposes its own workflow.
  • The context-saving benefits depend on how you use it.

A zero-setup grounded web-research MCP server (and Pi extension) — a good example of an MCP server delivered as a ready-to-install Pi package rather than something you host yourself.

Terminal window
pi install npm:@black-knight.dev/emet

Benefits

  • Zero setup — no separate server to provision.
  • Combines MCP plumbing with a useful built-in capability (research).

Drawbacks

  • Scoped to its research use case, not a general MCP client.
  • Want to connect arbitrary MCP servers: use pi-mcp-adapter.
  • Care most about context window cost: try context-mode.
  • Just want grounded web research without standing up a server: emet.