MCP Rubber Duck is an MCP server that brings rubber duck debugging to the AI age. Instead of explaining your problem to one plastic duck, you query multiple LLMs in parallel and get different perspectives right inside your editor.

What it does:

- Point it at OpenAI, Gemini, Groq, Ollama, LM Studio, or any OpenAI-compatible API

- Mix local and cloud models in the same query

- Compare answers side by side with interactive HTML panels

- Have models vote on the best approach with reasoning and confidence scores

- Run Oxford-style debates between models, with a judge picking the winner

- Maintain conversations with context across messages

Works with: Claude Desktop, Cursor, VS Code, and CLI agents (Claude Code, Codex, Gemini CLI, Grok, Aider).

Install: `npx mcp-rubber-duck` or Docker.

136 GitHub stars. MIT licensed.