think-mcp-server

Name: think-mcp-server
Author: MCP-Mirror

@MCP-Mirror

Repository

About think-mcp-server

Mirror of

#think-mcp-server #mcp-server #ai-assistant

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "marcopesani_think-mcp-server": {
      "command": "docker",
      "args": [
        "build",
        "-t",
        "think-mcp-server",
        "."
      ]
    }
  }
}

Tools

Enables Claude to think about a problem or analyze information

Overview

What is think-mcp-server?

think-mcp-server is a minimal Model Context Protocol (MCP) server that provides Claude AI models with the "think" tool capability. Based on Anthropic’s research published on March 20, 2025, the "think" tool gives Claude a dedicated reasoning step during response generation to improve complex, multi‑step tool‑use tasks.

How to use think-mcp-server?

Install globally via npm install -g think-mcp-server or run with npx -y think-mcp-server. For Claude Desktop, add the server configuration to claude_desktop_config.json (MacOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%/Claude/). Similar configuration steps are provided for Cline, Cursor, and Docker. Once configured, include instructions in your prompts to tell Claude when and how to use the “think” tool.

Key features of think-mcp-server

Implements one tool: think with a required thought string parameter.
Communicates over stdio using the MCP protocol.
Returns structured responses that AI assistants can process.
Logs thinking steps without modifying the external environment.
Based on Anthropic’s “think” tool research (March 2025).

Use cases of think-mcp-server

Analyzing previous tool call outputs before acting.
Following detailed guidelines in policy‑heavy environments.
Sequential decision making where each action builds on previous ones.

FAQ from think-mcp-server

What does the “think” tool do?

It allows Claude to pause during response generation and add an explicit thinking step—with its own designated space—to reason about whether all necessary information is available before proceeding.

How do I install think-mcp-server?

You can install it globally via npm (npm install -g think-mcp-server) or run it on demand with npx -y think-mcp-server. Docker build and run instructions are also provided.

How do I configure it for Claude Desktop?

Add the following JSON to your claude_desktop_config.json file (located in ~/Library/Application Support/Claude/ on MacOS or %APPDATA%/Claude/ on Windows): {"mcpServers":{"think":{"command":"npx","args":["-y","think-mcp-server"]}}}.

What performance improvements does the “think” tool provide?

Anthropic’s evaluations showed a 54% relative improvement on the τ‑Bench airline domain (pass^1 metric: 0.570 vs 0.370 baseline) and a 1.6% average improvement on SWE‑bench, along with enhanced consistency across trials.

What transport does think-mcp-server use?

The server runs as a standalone process and communicates over stdio using the Model Context Protocol. No authentication or additional transport is mentioned.