Zen MCP acts like a digital project manager for AI, allowing multiple models to work together as a unified team rather than in isolation. Instead of being limited to the perspective of just one AI, developers can use this tool to bring in various "experts"—such as Gemini for its massive memory or GPT for its reasoning—to collaborate on the same coding task. It essentially turns a standard command-line interface into a hub where different AI brains can talk to each other to solve complex problems more effectively. On a technical level, this server facilitates seamless multi-model orchestration with true conversation continuity. This means that context is preserved across different tools and models; if one model suggests a code change, the next model in the workflow actually understands the reasoning behind that suggestion. It includes specialized built-in workflows for systematic planning, deep-dive code reviews, and pre-commit validations, ensuring that the AI follows a rigorous process rather than jumping to conclusions. For advanced development workflows, the tool introduces a "clink" (CLI-to-Link) feature that can spawn isolated subagents. This allows a primary AI session to delegate heavy tasks—like auditing a massive security module—to a separate sub-CLI, keeping the main workspace clean while receiving only the final, relevant results. By leveraging the specific strengths of various providers like OpenRouter, Anthropic, and local Ollama instances, Zen MCP bypasses standard token limits and provides a way to "revive" lost context, making it an essential bridge for high-level automated engineering.
Category: AI & LLM Tooling
Tags: agent-framework, llm-routing, multi-agent, orchestration, workflow-automation
bash git clone https://github.com/BeehiveInnovations/zen-mcp-server.git cd zen-mcp-server 2. Run the setup script: bash ./run-server.sh This script handles setup, configuration, and auto-configures clients like Claude Desktop, Claude Code, Gemini CLI, Codex CLI, and Qwen CLI. Method B: Instant Setup with uvx Add the configuration directly to your MCP settings file (see below). ---~/.claude/settings.json or .mcp.json). Ensure you add your specific API keys in the env section. json { "mcpServers": { "zen": { "command": "bash", "args": [ "-c", "for p in $(which uvx 2>/dev/null) $HOME/.local/bin/uvx /opt/homebrew/bin/uvx /usr/local/bin/uvx uvx; do [ -x \"$p\" ] && exec \"$p\" --from git+https://github.com/BeehiveInnovations/zen-mcp-server.git zen-mcp-server; done; echo 'uvx not found' >&2; exit 1" ], "env": { "PATH": "/usr/local/bin:/usr/bin:/bin:/opt/homebrew/bin:~/.local/bin", "GEMINI_API_KEY": "your-key-here", "OPENROUTER_API_KEY": "your-key-here", "DISABLED_TOOLS": "analyze,refactor,testgen,secaudit,docgen,tracer", "DEFAULT_MODEL": "auto" } } } } Environment Variables: * GEMINI_API_KEY, OPENROUTER_API_KEY, etc.: Your AI provider keys. * DISABLED_TOOLS: Comma-separated list of tools to disable. * DEFAULT_MODEL: The model to use by default (e.g., "auto"). ---clink: Connects external AI CLIs (Gemini CLI, Claude Code, etc.) directly into the workflow and manages CLI subagents. * codereview: Performs multi-pass code analysis using specified models. * planner: Breaks down complex tasks or major refactors into simpler steps. * precommit: Final review of code changes before committing. * analyze: Code analysis (listed in configurable tools). * refactor: Code refactoring (listed in configurable tools). * testgen: Test generation (listed in configurable tools). * secaudit: Security auditing (listed in configurable tools). * docgen: Documentation generation (listed in configurable tools). * tracer: Execution tracing/debugging (listed in configurable tools). ---"Use zen to analyze this code for security issues with gemini pro" * Multi-Model Debugging: "Debug this error with o3 and then get flash to suggest optimizations" * Strategy Planning: "Plan the migration strategy with zen, get consensus from multiple models" * CLI Linkage: "clink with cli_name=\"gemini\" role=\"planner\" to draft a phased rollout plan" * Full Workflow: "Perform a codereview using gemini pro and o3 and use planner to generate a detailed plan, implement the fixes and do a final precommit check by continuing from the previous codereview"deep dive), the user can identify vulnerabilities that a single pass might miss. Zen handles the "hand-off," ensuring that if one model finds a potential leak, the next model validates it within the same conversation thread. Example: 1. Run: Perform a security audit using gemini-pro and o3 on the /auth directory. 2. Claude (orchestrator) walks the directory and collects files. 3. Gemini Pro performs a deep analysis and flags a potential SQL injection. 4. O3 reviews Gemini's finding to confirm if it’s a true positive. 5. Zen returns a unified report with a "high confidence" rating because both models agreed on the risk.
Use gemini-pro to analyze the entire repository architecture and explain how the data flows from the UI to the database. 2. Gemini ingests the massive context that would typically crash a standard MCP session. 3. Zen feeds the architectural summary back to Claude. 4. Even if Claude’s context resets, you can ask Zen to "revive" the session by having Gemini provide a briefing of previous findings to Claude.clink (CLI-Link) tool, Zen can spawn a "Subagent." This creates…Part of MCP Servers
Zen MCP, also known as PAL MCP, is a Model Context Protocol server that acts as a multi-model orchestration layer. It allows command-line AI tools and code editors to delegate tasks to different AI providers, such as OpenAI, Gemini, and Ollama, while preserving conversational context across steps.
Zen MCP works with any client supporting the Model Context Protocol, including Claude Code, Claude Desktop, Cursor, Gemini CLI, Codex CLI, Qwen Code CLI, and the Claude Dev VS Code extension.
Zen MCP enables multi-model consensus, systematic debugging, structured code refactoring, pre-commit validation, and isolated subagent execution. It provides built-in tools like planner, codereview, precommit, and clink to coordinate tasks across several AI providers without losing conversation state.
The clink tool connects external AI CLIs into your primary workflow. It lets a primary assistant spawn isolated subagents in fresh contexts to handle intensive tasks, such as deep file reviews or documentation lookups, returning only the summary back to the main workspace.