Happy slice of mcp pizzaMCP.pizza

think-mcp

MCP.Pizza Chef: Rai220

There is only one tool in the standard setup, and all it does is hand back whatever note the assistant wrote into it. That sounds pointless, but the forced pause is the point, and it is what Anthropic's own published testing found makes an assistant more careful on long multi-step jobs. Nothing needs signing up for, and nothing on your computer is read or changed. An optional advanced mode adds planning, self-criticism and web search, and only the search part needs a free Tavily account.

Other
Web/Research

Use This MCP server To

Make my assistant double-check its plan before it books anything Get it to re-read a tool's result before acting on it Have it list the rules that apply before answering See the reasoning it used, written out step by step Slow a long job down so fewer mistakes slip through

README

Think MCP Tool

Think MCP is an implementation of an MCP (Model Context Protocol) server that provides a "think" tool for structured reasoning in agentic AI workflows. This project is inspired by the Anthropic engineering article: The "think" tool: Enabling Claude to stop and think in complex tool use situations.

According to the referenced article, adding the think tool can lead to improved evaluation metrics by enabling reasoning capabilities even in models that do not natively possess advanced reasoning skills.

alt text

Think Tool MCP server

What is the "think" tool?

The "think" tool allows an AI agent to pause and record an explicit thought during complex reasoning or multi-step tool use. It does not change the environment or database, but appends the thought to the log, helping the agent process information, backtrack, or comply with detailed policies.

This approach is especially useful for:

  • Tool output analysis (processing results of previous tool calls)
  • Policy-heavy environments (verifying compliance with guidelines)
  • Sequential decision making (where each step builds on previous ones)

Features

  • Implements the "think" tool as described in Anthropic's research
  • Minimal, standards-based MCP server using mcp[cli]
  • Ready for integration with Claude or other agentic LLMs

Usage

MCP server configuration

Add this MCP server to your facorite agent.

"mcpServers": {
    "think-mcp": {
        "command": "uvx",
        "args": ["think-mcp"],
        "enabled": true
    }
}

Tool definition

The "think" tool is defined as:

  • Input: thought (string) — A thought to think about.
  • Behavior: Appends the thought to the log for structured reasoning.

Advanced mode

Adds aditional tools for your agent:

  • criticize
  • plan
  • search
"mcpServers": {
    "think-mcp": {
        "command": "uvx",
        "args": ["think-mcp", "--advanced"],
        "enabled": true,
        "env": {
            "TAVILY_API_KEY": ... YOUR TAVILY API KEY HERE ...
        }
    }
}

Reference

License

MIT License — see LICENSE

think-mcp FAQ

Do I need a key or an account?
Not for the normal setup, which is completely free and needs no sign-up. Only the optional advanced mode's web search wants a Tavily key, and Tavily has a free tier.
How hard is setup?
Easy. You paste a three-line block into your assistant's settings and it fetches the program itself the first time.
Does it touch anything on my computer?
No. It reads no files, changes nothing, and reaches out to the internet only if you switch on the optional search tool.
Does it store the thoughts anywhere?
No, despite its own wording about writing to a log. The text is handed straight back into the conversation and nothing is saved to disk.
Does it actually make a difference?
It depends on the model. The research it is built on reports clear gains for models that do not already reason at length on their own, so a model that already thinks step by step may gain little.
Is there a catch with the advanced mode?
Yes, one. Switching it on makes the program print a plain-text startup line onto the same channel it uses to talk to your assistant, which some apps report as a failure to start. If it connects fine until you add that switch, this is why.
Can I use this to make my assistant stick to a refund policy?
Yes, and that is close to the intended use. It also ships a ready-made instruction block that tells the assistant when to pause and what to check.
Which apps does it work in?
Any assistant that can start a local command, which covers Claude Desktop, Cursor, Cline and most others.