Skip to content

MCP Servers

LazyAI maintains a canonical MCP catalog under .ai/mcp.json and compiles it into each tool's native config format (see MCP Integration for the compile flow and per-tool output files). This page documents the servers that ship in the catalog.

The authoritative source is packages/cli/library/mcp/catalog.json (schema version 1.0). Each server records its transport, the tools it exposes, whether it requires a separate install, and an optional setup hint. Servers also declare a preferred_interface and cli_equivalent so the agent can choose the cheaper CLI path for bulk work where one exists.

Enabling servers

# At init
lazyai-cli init --enable-servers filesystem,ai-memory

# Later: edit .ai/mcp.json (set "enabled": true on a server) then recompile
lazyai-cli compile

LazyAI does not create or manage .env files. When an enabled server declares environment variables, lazyai-cli prints shell-ready export NAME="" guidance; supply real values through your shell, a secret manager, or a local env file per your project policy.

Cataloged servers

Server Transport Requires install Preferred interface Purpose
ai-memory HTTP (http://127.0.0.1:49374/mcp) Yes hybrid Long-term project memory, handoffs, and durable annotations
filesystem stdio (npx @modelcontextprotocol/server-filesystem) No cli Local filesystem read/write access
ripgrep stdio (npx mcp-ripgrep) No cli Fast code search via ripgrep
codegraph stdio (codegraph serve --mcp) Yes hybrid Code knowledge graph for semantic exploration
obsidian stdio (ob mcp) Yes cli-first Obsidian vault read/write for persistent knowledge

ai-memory

  • Tools: memory_query, memory_recent, memory_handoff_accept, memory_write_page, memory_auto_improve, memory_consolidate
  • CLI equivalent: ai-memory
  • Install: curl -fsSL https://raw.githubusercontent.com/akitaonrails/ai-memory/main/bin/ai-memory -o ~/.local/bin/ai-memory && chmod +x ~/.local/bin/ai-memory
  • Setup: Start the ai-memory server, then point clients at http://127.0.0.1:49374/mcp or a remote /mcp endpoint. Add an Authorization header when bearer auth is enabled.

filesystem

  • Tools: read_file, write_file, list_directory, search_files
  • CLI equivalent: native shell / lazyai-cli file ops
  • Install: none (runs via npx -y @modelcontextprotocol/server-filesystem .)
  • Enabled by default; prefer native shell/CLI for bulk file work.

ripgrep

  • Tools: search, list-files
  • CLI equivalent: rg
  • Install: none (runs via npx -y mcp-ripgrep)

codegraph

  • Tools: codegraph_context, codegraph_search, codegraph_files
  • CLI equivalent: codegraph
  • Install: bun install -g codegraph
  • Setup: Initialize the repository index with codegraph init.

obsidian

  • Tools: read_note, write_note, search_notes, list_notes
  • CLI equivalent: ob
  • Install: npm install -g obsidian-cli
  • Setup: Ensure the Obsidian "Advanced URI" plugin is installed and enabled.

Per-tool compilation

The catalog compiles into each tool's native MCP config. See MCP Integration and Tool Outputs for the exact output files per tool (for example opencode.json for OpenCode, .mcp.json for Claude Code, .vscode/mcp.json for Copilot).

L1 vs L3 validation

lazyai-cli server doctor performs L1 config checks only: it verifies that the server entry exists in .ai/mcp.json, is enabled, and is present in each per-tool compiled MCP config file.

L3 stdio handshake (spawning the server process and performing a tools/list JSON-RPC exchange) is not performed. The Go binary does not bundle an MCP client library for the handshake protocol. A TypeScript wrapper using @modelcontextprotocol/sdk could perform L3 checks; this is future work.

The server doctor output includes a stdio handshake check that is always skipped, with a message explaining the limitation.