Pal MCP Server — Multi-Model AI Gateway for Claude Code

What it is

Pal MCP Server is a Model Context Protocol gateway that connects Claude Code to multiple LLM providers through a single configuration. It supports Gemini, OpenAI, Azure OpenAI, X.AI/Grok, OpenRouter, DIAL, and Ollama, letting you route prompts to the best model for each task without switching tools.

The server is designed for developers who want to leverage model diversity from within their existing Claude Code workflow. Instead of context-switching between CLIs, you send requests through Pal and let it handle provider routing, conversation continuity, and response formatting.

How it saves time or tokens

Pal reduces token waste by routing tasks to the most cost-effective model. A simple code formatting request can go to a smaller model, while complex architecture decisions stay with Claude. The CLI-to-CLI bridge feature (called 'clink') maintains conversation context across model switches, so you do not lose context when switching providers mid-session. This avoids re-prompting and the token overhead that comes with it.

How to use

1. Install prerequisites: Python 3.10+, Git, and uv (pip install uv).
2. Add the Pal server to your .mcp.json configuration:

{
  "mcpServers": {
    "pal": {
      "command": "bash",
      "args": ["-c", "uvx --from git+https://github.com/BeehiveInnovations/pal-mcp-server.git pal-mcp-server"],
      "env": {
        "GEMINI_API_KEY": "your-gemini-key",
        "DEFAULT_MODEL": "auto"
      }
    }
  }
}

1. Restart Claude Code. The tools chat, thinkdeep, planner, and consensus become available for multi-model interaction.

Example

Send a planning request through Pal to get perspectives from multiple models:

# In Claude Code, use the planner tool
'Plan a microservices migration for a monolithic Node.js app'
# Pal routes sub-tasks to different models and merges results

The consensus tool queries multiple providers with the same prompt and synthesizes agreement points, useful for architecture decisions where you want diverse AI perspectives.

Related on TokRepo

• AI Gateway Providers — Compare gateway solutions like LiteLLM, OpenRouter, and Portkey alongside Pal
• MCP Integrations — Browse other MCP server configurations for extending Claude Code

Common pitfalls

• Setting DEFAULT_MODEL to a specific model instead of 'auto' defeats the routing benefit. Keep it on 'auto' unless you have a reason to pin.
• Each provider needs its own API key in the env block. Missing keys cause silent failures on routes to that provider.
• Conversation continuity via clink requires all participating models to be available. If one provider is down, the context chain breaks.
• Always check the official documentation for the latest version-specific changes and migration guides before upgrading in production environments.