multi_mcp

Multi-MCP is a multi-model AI orchestration MCP server for automating code review and LLM-based analysis. It integrates Claude Code CLI and OpenCode, and can orchestrate multiple AI models (OpenAI GPT, Anthropic Claude, Google Gemini) for code quality checks, security analysis (OWASP Top 10), and multi-agent consensus. The tool is built on the Model Context Protocol (MCP), enabling Python developers and DevOps teams to automate code review directly in their development workflows using AI-driven insights.

Key capabilities:

• Code review: A systematic workflow including OWASP Top 10 security checks and performance analysis.
• Chat: Interactive development assistance with repository context awareness.
• Comparison: Parallel multi-model analysis for architectural decisions.
• Debate: A multi-agent consensus workflow (independent answers + critique).
• Multi-model support: OpenAI GPT, Anthropic Claude, Google Gemini, and OpenRouter.
• CLI and API models: Can mix CLI-based models (Gemini CLI, Codex CLI) and API models.
• Model aliases: Can use short names such as mini, sonnet, gemini.
• Threads: Maintain context across multi-step reviews.

Usage: Multi-MCP acts as an MCP server, connected by Claude Code or OpenCode, providing AI-driven code analysis tools. After installing the MCP server and configuring AI model API keys, it can be automatically integrated into Claude Code or OpenCode via make install. Use natural language to invoke tools (for example, "multi codereview this file") to get results from multiple AI models in parallel.

Fast multi-model analysis features:

• Parallel execution: 3 models in about 10 seconds (compared with about 30 seconds for sequential execution).
• Asynchronous architecture: Non-blocking Python asyncio.
• Conversation threads: Maintain context across multi-step reviews.
• Low latency: Response time equals the time of the slowest model, not the sum of all model times.

Environment requirements: At least one provider's API key (OpenAI, Anthropic, Google, or OpenRouter).

Manual integration: If you do not run make install, you can add the configuration to Claude Code (~/.claude.json) or OpenCode (~/.opencode/opencode.json), replacing /path/to/multi_mcp with the actual clone path.

Environment configuration (API keys and settings): Multi-MCP loads settings from the .env file in the following order (priority from high to low):

1. Environment variables (already set in the shell)
2. Project .env (current directory or project root)
3. User .env (~/.multi_mcp/.env) - fallback location for pip installation Edit .env to fill in API keys.

Model configuration (adding custom models): Models are defined in YAML configuration files (user configuration takes precedence):

• Package defaults: multi_mcp/config/config.yaml (bundled with the package)
• User override: ~/.multi_mcp/config.yaml (optional, takes precedence) To add your own models, create ~/.multi_mcp/config.yaml (refer to the config.yaml and config.override.example.yaml examples). Merge behavior:
• New models are added alongside the package defaults
• Existing models are merged (your settings override the package defaults)
• Aliases can be "transferred" from package models to custom models

After installation into an MCP client (Claude Code or OpenCode), the following commands can be used:

• Chat: Interactive development assistance
• Code review: Analyze code using a specific model
• Comparison: Get multiple perspectives (using the default model)
• Debate: In-depth analysis with critique

Enable the allowlist: Edit ~/.claude/settings.json and add the corresponding line under permissions.allow so that Claude Code can use Multi MCP without user permission.

Use short aliases instead of full model names: Run multi:models to view all available models and aliases.

CLI models: Multi-MCP can execute CLI-based AI models (such as Gemini CLI, Codex CLI, or Claude CLI) as well as API models. CLI models run as subprocesses and work seamlessly with all existing tools.

• Use models with full tool access (file operations, shell commands)
• Mix API and CLI models in comparison and debate workflows
• Leverage local CLIs without API overhead Built-in CLI models:
• gemini-cli (alias: gem-cli) — Gemini CLI, auto-edit mode
• codex-cli (alias: cx-cli) — Codex CLI, fully automatic mode
• claude-cli (alias: cl-cli) — Claude CLI, acceptEdits mode Adding custom CLI models: Add to ~/.multi_mcp/config.yaml (see model configuration). CLI models require the corresponding CLI tool to be installed.

CLI usage (experimental): Multi-MCP includes a standalone CLI for code review without an MCP client. Note: the CLI is experimental and under active development.

Troubleshooting:

• "No API key found": add at least one API key in the .env file; verify it is loaded: uv run python -c "from multi_mcp.settings import settings; print(settings.openai_api_key)"
• Integration tests failing: set the RUN_E2E=1 environment variable; verify the API key is valid and has sufficient quota; check logs/server.log for detailed logs.

FAQ:

• Are all three AI providers required? No, just one API key (OpenAI, Anthropic, or Google) is enough to get started.
• Do they truly run in parallel? Yes. When using the codereview, compare, or debate tools, all models execute concurrently via Python's asyncio.gather(), and the response time equals the time of the slowest model, not the sum of all response times.
• How many models can run simultaneously? There is no hard limit. In practice, 2-5 models work for most use cases. All tools use the configured default models (usually 2-3), but any number of models can be specified.

Contributing: Contributions are welcome; see CONTRIBUTING.md for development setup, testing guidelines, and the pull request process.

License: MIT License, see the LICENSE file for details.