PAL MCP
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
Install
git clone https://github.com/BeehiveInnovations/pal-mcp-server.git
cd pal-mcp-server
./run-server.shRequires Python 3.10+, Git and uv. The script configures clients automatically.
This is third-party code. Review the repository files before installing.
What it does
PAL, formerly Zen MCP, connects other models to your agent through OpenRouter, Gemini, OpenAI, Azure, X.AI, DIAL or Ollama. The agent can discuss a solution with a second model, gather opinions from several models, and run step-by-step code review, debugging or pre-commit checks while keeping a shared conversation context. The clink tool launches external CLIs such as Gemini CLI or Codex as subagents with separate context. Extra tools are off by default to save context.
Who it is for. For developers who want a second opinion from another model without leaving their agent.
Good fit when
- You want an independent code review from another model
- A hard bug where comparing hypotheses from different models helps
- You need to hand a large analysis to a long-context model
Not a fit when
- You have no provider keys and no local models
- Code must not be sent to several external providers
Example request
Review the auth module together with gemini pro and compile a combined issue listLimitations
Needs keys for at least one provider or a local Ollama. OpenAI, Gemini and X.AI APIs do not work from Russia without a VPN. Each call spends tokens with external providers. Tool descriptions take context space, so some are disabled. Recent commit activity is sparse.
How to disable. Remove the pal entry from ~/.claude/settings.json, .mcp.json or another client's config.
MCP
- Transport
- stdio
- Authentication
- API key
| Environment variables | |
|---|---|
| GEMINI_API_KEY secret | Gemini key; at least one provider key is required |
| DISABLED_TOOLS | Comma-separated list of disabled tools |
| DEFAULT_MODEL | Default model, auto for automatic selection |
Security check
- Sends code and context to external model providers
- Stores API keys for several providers
README in short
The README explains an agent working with several models in one conversation thread and walks through a multi-step review example. Install by cloning and running run-server.sh or via uvx in an MCP config snippet. It covers providers, grouped tools, which are on by default and how to enable more. It also introduces clink for running other CLIs as subagents.
FAQ
Which tools are on by default?
chat, thinkdeep, planner, consensus, codereview, precommit, debug, apilookup and challenge. Others are enabled via DISABLED_TOOLS.
Can I use it without cloud keys?
Yes, by connecting local models through Ollama.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent