PAL MCP Server (formerly Zen MCP)

Use it when

• Your coding CLI is locked to one vendor and you want second opinions, consensus debates, or code reviews from Gemini, OpenAI, Grok, or local Ollama models inside the same conversation thread
• Your assistant's context window overflows on large codebases or resets mid-task and you need to delegate to larger-context models and revive the conversation with full continuity

What it solves

Not the fit when

• Remote providers require API credentials and incur their own usage charges; a local Ollama configuration is the no-provider-cost option
• Prompts, code excerpts, and conversation context sent to a remote model cross that provider's privacy and retention boundary
• The clink bridge launches external agent CLIs with their configured filesystem, shell, web, and MCP capabilities; constrain their working directory and permissions before delegating sensitive or consequential tasks
• Analyze, refactor, testgen, secaudit, docgen, and tracer are disabled by default and must be explicitly enabled
• Every enabled MCP tool consumes client context even when unused
• hosting or serving models itself
• long-term agent memory store
• browser or ui automation

Install

git clone https://github.com/BeehiveInnovations/pal-mcp-server.git && cd pal-mcp-server && ./run-server.sh (or add the uvx-based pal entry to ~/.claude/settings.json or .mcp.json with your API keys)

Invoke

From your CLI, prompt the pal tools, e.g. "Use pal to analyze this code for security issues with gemini pro", "Use consensus with gpt-5 and gemini-pro to decide ...", or "clink with codex codereviewer to audit auth module"

Alternatives

No reviewed alternatives recorded yet.