🧠 ThinkingCap

A multi-agent research MCP server that runs multiple LLM providers in parallel and synthesizes their responses. Built on the Model Context Protocol for seamless integration with Claude Desktop, Cursor, and other MCP-compatible tools.

🌟 Features

• 🔀 Multi-Agent Research: Deploy multiple AI agents simultaneously for comprehensive analysis
• 🎯 Multi-Provider Support: OpenAI, Anthropic, xAI, Google, OpenRouter, Groq, Cerebras, Fireworks, Together, DeepInfra
• ⚡ Parallel Execution: All agents run concurrently for maximum speed
• 🔄 Intelligent Synthesis: Combines multiple perspectives into unified, comprehensive answers
• 🔍 Built-in Web Search: DuckDuckGo search integration (no API key required)
• 🔌 MCP Native: Works with any MCP-compatible client via npx

🚀 Quick Start

Installation

No installation required! Just add to your MCP client configuration.

Configuration

Add the following to your MCP client configuration (e.g., ~/.cursor/mcp.json):

{
  "mcpServers": {
    "thinkingcap": {
      "command": "npx",
      "args": [
        "-y",
        "thinkingcap",
        "openrouter:openai/gpt-4o",
        "groq:llama-3.3-70b-versatile",
        "cerebras:llama-3.3-70b",
        "xai:grok-2-latest"
      ]
    }
  }
}

Customizing Agents

You can specify any combination of providers and models as arguments:

"args": [
  "-y",
  "thinkingcap",
  "anthropic:claude-sonnet-4-20250514",
  "openai:gpt-4o",
  "google:gemini-2.0-flash"
]

📋 Supported Providers

| Provider | Env Variable | Default Model | Example | |----------|--------------|---------------|---------| | openai | OPENAI_API_KEY | gpt-4o | openai:gpt-4o-mini | | openrouter | OPENROUTER_API_KEY | openai/gpt-4o | openrouter:anthropic/claude-3.5-sonnet | | groq | GROQ_API_KEY | llama-3.3-70b-versatile | groq | | cerebras | CEREBRAS_API_KEY | llama-3.3-70b | cerebras | | xai | XAI_API_KEY | grok-2-latest | xai:grok-2-latest | | anthropic | ANTHROPIC_API_KEY | claude-sonnet-4-20250514 | anthropic | | google | GOOGLE_API_KEY | gemini-2.0-flash | google:gemini-1.5-pro | | fireworks | FIREWORKS_API_KEY | llama-v3p3-70b-instruct | fireworks | | together | TOGETHER_API_KEY | Llama-3.3-70B-Instruct-Turbo | together | | deepinfra | DEEPINFRA_API_KEY | Llama-3.3-70B-Instruct | deepinfra |

🔑 Environment Variables

API keys are read from environment variables. Add them to your ~/.bashrc or ~/.zshrc:

export OPENROUTER_API_KEY="sk-or-..."
export GROQ_API_KEY="gsk_..."
export CEREBRAS_API_KEY="..."
export XAI_API_KEY="..."
# etc.

🛠️ How It Works

1. Query Decomposition: Your research query is broken into multiple specialized questions
2. Parallel Execution: Each agent (provider/model combo) researches a different angle
3. Web Search: Each agent performs web searches to gather current information
4. Synthesis: All agent responses are combined into one comprehensive answer

🔥 OpenRouter Fireworks Routing

When using OpenRouter, requests are automatically routed to Fireworks as the preferred provider with fallbacks enabled for maximum reliability.

📝 License

MIT License

🙏 Acknowledgments

• Built on the Model Context Protocol
• Inspired by multi-agent AI research systems