multi_mcp CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

Multi-MCP is a multi-model AI orchestration server that provides advanced code analysis capabilities through the Model Context Protocol (MCP). It orchestrates multiple LLM providers via LiteLLM to deliver systematic code review and analysis tools.

The server is built with FastMCP and uses a streamlined workflow architecture optimized for fast, cost-effective analysis with models like gpt-5-mini.

Current Status

Production Ready ✅

• Unit Tests: ✅ 571 unit tests (~2s) - mocked, no API keys needed (includes CLI tests for cli.py and cli_executor.py)
• Integration Tests: ✅ 96 tests (~8-10min) - real API calls, run locally before PR
• Total Coverage: ✅ 667 tests total (~93% code coverage)
• Cross-platform: ✅ unit suite green on Linux, macOS, and Windows (CI matrix)
• Model Config: YAML-based model configuration with aliases and use-case defaults
• Logging: MCP tool request/response logging enabled
• Implementation: Checklist-based workflow with expert validation enabled
• File Limit Enforcement: ✅ settings.max_files_per_review is enforced

Development Commands

# Type checking (required before commits)
uv run pyright

# Linting and formatting (required before commits)
uv run ruff check .
uv run ruff format .

# Run all unit tests (571 tests, ~2s, all passing ✅)
uv run pytest tests/unit/ -v

# Run integration tests (96 tests, ~5-7min with parallel, all passing ✅)
# Note: Requires real API keys (OPENAI_API_KEY, etc.)
# CLI tests will skip gracefully if CLIs not installed
RUN_E2E=1 uv run pytest tests/integration/ -n auto -v

# Or run sequentially (slower, ~15min)
RUN_E2E=1 uv run pytest tests/integration/ -v

# Run all tests (667 total)
RUN_E2E=1 uv run pytest tests/ -v

# Run the MCP server
./scripts/run_server.sh
# or: uv run python multi_mcp/server.py

# View MCP logs (request/response)
ls -lh logs/*.mcp.json
cat logs/*.mcp.json | jq .