Think MCP Server

What is Think MCP Server?

Think MCP Server uses Groq’s API to call LLMs that expose raw chain‑of‑thought tokens from Qwen’s qwq model. It is designed to be used as an MCP tool by AI agents (e.g., Claude) to improve performance on complex tool‑use situations, following the approach described in Anthropic’s “think” tool article.

How to use Think MCP Server?

Clone the repository, run uv sync to install dependencies, obtain a Groq API key, and update your MCP client configuration with the path to the repository and the GROQ_API_KEY environment variable. Then instruct your AI agent to call the chain_of_thought tool on every request—typically by adding an XML rule that defines when and how to use it as a scratchpad for reasoning.

Key features of Think MCP Server

• Exposes chain‑of‑thought tokens via Groq’s API from Qwen’s qwq model.
• Integrates as a standard MCP tool for any compatible AI client.
• Provides an external scratchpad for structured reasoning and planning.
• Shown to increase performance on SWE Bench and similar benchmarks.

Use cases of Think MCP Server

• Improving an AI agent’s performance on software engineering tasks (e.g., SWE Bench).
• Acting as a step‑by‑step reasoning aid for complex multi‑step tool calls.
• Enforcing policy compliance by having the agent verify rules before acting.

FAQ from Think MCP Server

What does Think MCP Server actually do?

It calls an LLM (via Groq) that returns raw chain‑of‑thought tokens, which the AI agent can use as a scratchpad to reason through a task before responding or taking further tool actions.

What dependencies does it require?

You need Python, uv for dependency management, a Groq API key, and network access to the Groq API. No additional runtime or database is required.

How do I configure the server in my MCP client?

Update your MCP configuration JSON with the command uv run src/server.py from the repository directory and set the GROQ_API_KEY environment variable to your key.

How do I make my AI agent call the tool?

Add an XML instruction in your agent’s rules (or system prompt) telling it to call the chain_of_thought tool on every user request and use it as a structured scratchpad for planning and verification.

Where does my data go when using the tool?

The server sends prompts to the Groq API to generate chain‑of‑thought output; no persistent storage of user data is described in the repository.