Rubber Duck

MCP Rubber Duck

An MCP (Model Context Protocol) server that acts as a bridge to query multiple LLMs -- both OpenAI-compatible HTTP APIs and CLI coding agents. Just like rubber duck debugging, explain your problems to various AI "ducks" and get different perspectives!

Why direct provider integration? MCP's sampling primitive -- a server borrowing the *host's* model -- was deprecated in the 2026-07-28 spec RC in favor of servers integrating directly with LLM provider APIs. Rubber Duck has always worked this way (it brings its own ducks), so it's aligned with where the protocol is heading -- no migration required.

Features

• Universal OpenAI Compatibility -- Works with any OpenAI-compatible API endpoint
• CLI Agent Support -- Use CLI coding agents (Claude Code, Codex, Gemini CLI, Grok, Aider) as ducks
• Multiple Ducks -- Configure and query multiple LLM providers simultaneously
• Conversation Management -- Maintain context across multiple messages
• Duck Council -- Get responses from all your configured LLMs at once
• Consensus Voting -- Multi-duck voting with reasoning and confidence scores
• LLM-as-Judge -- Have ducks evaluate and rank each other's responses
• Iterative Refinement -- Two ducks collaboratively improve responses
• Structured Debates -- Oxford, Socratic, and adversarial debate formats
• MCP Prompts -- 8 reusable prompt templates for multi-LLM workflows
• Vision Input -- Send images alongside prompts to vision-capable models (docs)
• Automatic Failover -- Falls back to other providers if primary fails
• Health Monitoring -- Real-time health checks for all providers
• Usage Tracking -- Track requests, tokens, and estimated costs per provider
• MCP Bridge -- Connect ducks to other MCP servers for extended functionality (docs)
• Guardrails -- Pluggable safety layer with rate limiting, token limits, pattern blocking, and PII redaction (docs)
• Granular Security -- Per-server approval controls with session-based approvals
• Interactive UIs -- Rich HTML panels for compare, vote, debate, and usage tools (via MCP Apps)
• Tool Annotations -- MCP-compliant hints for tool behavior (read-only, destructive, etc.)
• Structured Output -- outputSchema on tools returning structured JSON for client-side validation (Cursor, VS Code/Copilot)
• Spec-Aligned by Design -- connects directly to provider APIs, the path the MCP 2026-07-28 spec recommends now that server-side sampling is deprecated (SEP-2577)

Supported Providers

HTTP Providers (OpenAI-compatible API)

Any provider with an OpenAI-compatible API endpoint, including:

• OpenAI
• Google Gemini
• Anthropic (via OpenAI-compatible endpoints)
• Groq (fast inference for open-weight models)
• Together AI (broad open-weight model catalog)
• Perplexity (online models with web search)
• Azure OpenAI, Ollama, LM Studio, Custom

CLI Providers (Coding Agents)

Command-line coding agents that run as local processes:

• Claude Code (claude) -- Codex (codex) -- Gemini CLI (gemini) -- Grok CLI (grok) -- Aider (aider) -- Custom

See CLI Providers for full setup and configuration.

Quick Start

# Install globally

npm install -g mcp-rubber-duck

Or use npx directly in Claude Desktop config

npx mcp-rubber-duck

Using Claude Desktop? Jump to Claude Desktop Configuration. Using Cursor, VS Code, Windsurf, or another tool? See the Setup Guide.

Installation

Prerequisites

• Node.js 20 or higher
• npm or yarn
• At least one API key for an HTTP provider, or a CLI coding agent installed locally

Install from NPM

npm install -g mcp-rubber-duck

Install from Source

git clone https://github.com/nesquikm/mcp-rubber-duck.git

cd mcp-rubber-duck npm install npm run build npm start

Configuration

Create a .env file or config/config.json. Key environment variables:

Full reference: [Configuration docs](./docs/configuration.md)

Interactive UIs (MCP Apps)

Four tools -- compare_ducks, duck_vote, duck_debate, and get_usage_stats -- can render rich interactive HTML panels inside supported MCP clients via MCP Apps. Once this MCP server is configured in a supporting client, the UIs appear automatically -- no additional setup is required. Clients without MCP Apps support still receive the same plain text output (no functionality is lost). See the MCP Apps repo for an up-to-date list of supported clients.

Compare Ducks

Compare multiple model responses side-by-side, with latency indicators, token counts, model badges, and error states.

Duck Vote

Have multiple ducks vote on options, displayed as a visual vote tally with bar charts, consensus badge, winner card, confidence bars, and collapsible reasoning.

Duck Debate

Structured multi-round debate between ducks, shown as a round-by-round view with format badge, participant list, collapsible rounds, and synthesis section.

Usage Stats

Usage analytics with summary cards, provider breakdown with expandable rows, token distribution bars, and estimated costs.

Available Tools

Full reference with input schemas: [Tools docs](./docs/tools.md)

Available Prompts

Full reference with examples: [Prompts docs](./docs/prompts.md)

Development

npm run dev        # Development with watch mode

npm test # Run all tests npm run lint # ESLint npm run typecheck # Type check without emit

Documentation

Troubleshooting

Provider Not Working

1. Check API key is correctly set
2. Verify endpoint URL is correct
3. Run health check: list_ducks({ check_health: true })
4. Check logs for detailed error messages

OpenAI gpt-5.6 / gpt-6 Models Fail With the MCP Bridge

OpenAI's Chat Completions API rejects function tools for the gpt-5.6 and gpt-6 families (400 Function tools with reasoning_effort are not supported ... use /v1/responses). They work for plain calls, but with MCP_BRIDGE_ENABLED=true the ducks attach tools and those requests fail. Use gpt-5.4 (the default) or gpt-5.4-mini until Responses API support lands (#166).

Connection Issues

• For local providers (Ollama, LM Studio), ensure they're running
• Check firewall settings for local endpoints
• Verify network connectivity to cloud providers

Rate Limiting

• Configure failover to alternate providers
• Adjust max_retries and timeout settings
• See Guardrails for rate limiting configuration

Contributing

     __

<(o )___ ( ._> / `---' Quack! Ready to debug!

We love contributions! Whether you're fixing bugs, adding features, or teaching our ducks new tricks, we'd love to have you join the flock.

Check out our Contributing Guide to get started.

Quick start for contributors:

1. Fork the repository
2. Create a feature branch
3. Follow our [conventional co