PAL MCP (formerly Zen)

PAL (it used to be called Zen MCP) turns Claude into the lead of a small AI team. Mid-task, Claude can ask Gemini, OpenAI's models, Grok or any model on OpenRouter to check its thinking, review its code or argue the other side, then pull the answers together. The same conversation carries across models, so nobody starts from scratch.

The headline tools are consensus (several models debate a decision), codereview, precommit, debug and thinkdeep. Six more, like refactor and testgen, are switched off by default to save context, and you can turn them on in the settings.

It runs locally with Python and uv. You need at least one API key (Gemini, OpenAI or OpenRouter) or a local Ollama model. The easiest setup is the run-server.sh script, which configures Claude Code for you. Costs depend on which models you call, so watch your usage. On Windows it needs WSL2.

• Ask Gemini, GPT or another model for a second opinion on a plan or code
• Run a "consensus" debate across several models before a big decision
• Do a thorough code review and a pre-commit check of your changes
• Track down a stubborn bug step by step
• Hand a huge codebase to a model with a bigger context window
• Call other AI command-line tools, like Codex or Gemini CLI, as helpers

Open your terminal in the project folder and run:

claude mcp add pal -e GEMINI_API_KEY=YOUR_API_KEY -- uvx --from git+https://github.com/BeehiveInnovations/pal-mcp-server.git pal-mcp-server

Other ways to set it up:

# Recommended: the setup script installs PAL and configures Claude Code for you
git clone https://github.com/BeehiveInnovations/pal-mcp-server.git
cd pal-mcp-server
./run-server.sh

You'll be asked to sign in or paste an API key the first time you use it.

Developers who want more than one AI checking important work.

• Second opinions on architecture
• Multi-model code review
• Debugging hard problems
• Big-context analysis with Gemini