Mcp Rubber Duck

What is MCP Rubber Duck?

MCP Rubber Duck is an MCP server that bridges to multiple LLMs—both OpenAI-compatible HTTP APIs and CLI coding agents—inspired by rubber duck debugging. It lets you explain problems to various AI “ducks” and get different perspectives.

How to use MCP Rubber Duck?

Install globally via npm install -g mcp-rubber-duck or use npx directly. Configure API keys in a .env file or config/config.json. Invoke by running the server and using the provided tools (e.g., ask_duck, compare_ducks, duck_debate) from an MCP client. For Claude Desktop, see the dedicated setup guide.

Key features of MCP Rubber Duck

• Universal OpenAI API compatibility for any endpoint.
• CLI coding agent support (Claude Code, Codex, Gemini CLI, etc.).
• Multiple ducks queried simultaneously (Duck Council).
• Consensus voting, LLM-as-judge, and iterative refinement.
• Structured debates (Oxford, Socratic, adversarial).
• 8 reusable MCP prompt templates for multi-LLM workflows.
• Automatic failover, health monitoring, and usage tracking.
• MCP Bridge to connect ducks to other MCP servers.
• Pluggable guardrails (rate limiting, token limits, PII redaction).
• Interactive UIs for compare, vote, debate, and usage tools.

Use cases of MCP Rubber Duck

• Debugging code by explaining problems to multiple AI models.
• Getting multi-perspective analysis on a plan or design.
• Conducting structured debates between LLMs to explore trade-offs.
• Generating and refining responses collaboratively between two ducks.
• Evaluating and ranking LLM responses using another LLM as a judge.

FAQ from MCP Rubber Duck

How does MCP Rubber Duck differ from MCP’s sampling primitive?

MCP Rubber Duck uses direct LLM provider integration, which aligns with the 2026-07-28 spec where server-side sampling was deprecated. It brings its own ducks and does not require the host’s model.

What runtime dependencies are required?

Node.js 20 or higher, npm or yarn, and at least one API key for an HTTP provider or a local CLI coding agent.

Which providers and CLI agents are supported?

OpenAI, Google Gemini, Anthropic, Groq, Together AI, Perplexity, Anyscale, Azure OpenAI, Ollama, LM Studio, and custom OpenAI-compatible endpoints. CLI agents: Claude Code, Codex, Gemini CLI, Grok, Aider, and custom.

Where does data live? How are conversations managed?

Data is handled locally by the server; conversation history is maintained in memory and can be cleared via clear_conversations. API keys are stored in environment variables or config files.

What transports and authentication are used?

The server uses MCP stdio transport by default. Authentication is via API keys configured for each provider; no built-in auth for the server itself.