MCP Think Tool Server

MCP Think Tool Server is an MCP server that implements Anthropic's think tool to provide AI assistants like Claude with a dedicated scratchpad for structured reasoning. Developed for developers, prompt engineers, and AI practitioners who deploy LLMs on intricate workflows, the server exposes tools to record, retrieve, analyze, and clear chain-of-thought processing steps during problem-solving sessions. By offering explicit reasoning space, the server enables Claude to break down multifaceted problems, review policies and constraints before invoking actions, and verify inputs and intermediate tool outputs. It exposes four primary tools: think for capturing structured steps, get_thoughts for inspecting prior reflections, clear_thoughts for resetting state between tasks, and get_thought_stats for analyzing thinking patterns. This structured approach helps maintain logical consistency across sequential decision-making workflows, prevents hallucinated assumptions during multi-step tool calls, and allows the model to deliberate on compliance requirements before delivering final responses to users.

Category: AI & LLM Tooling

Tags: chain-of-thought, cognitive-processing, prompt-engineering, reasoning

Visit MCP Think Tool Server

How to install and configure MCP Think Tool Server

1. Install the server using Smithery by running npx -y @smithery/cli install @cgize/claude-mcp-think-tool --client claude. 2. Alternatively, install globally using npm: bash npm install -g @cgize/mcp-think-tool 3. Open your Claude Desktop configuration file located at %APPDATA%\Claude\claude_desktop_config.json on Windows or the corresponding path on macOS. 4. Add the server entry to the mcpServers block: json { "mcpServers": { "think-tool": { "command": "npx", "args": [ "-y", "@cgize/mcp-think-tool" ], "type": "stdio", "pollingInterval": 30000, "startupTimeout": 30000, "restartOnFailure": true } } } 5. If installed globally, set "command": "claude-mcp-think-tool" with empty "args": []. 6. Restart Claude Desktop.

What you can do with MCP Think Tool Server

• Evaluating multi-step tool outputs by creating an intermediate scratchpad to verify API responses before making subsequent tool calls. - Checking policy compliance against complex operational guidelines before delivering user-facing responses or triggering critical actions. - Decomposing intricate mathematical or logic problems into explicit sequential calculation steps to prevent compound reasoning errors. - Tracking and auditing AI reasoning processes across extended sessions using the retrieval and thought statistics inspection tools.

Key facts

• https://github.com/cgize/claude-mcp-think-tool
• AI & LLM Tooling
• chain-of-thought, cognitive-processing, prompt-engineering, reasoning

Part of MCP Servers

What does MCP Think Tool Server do?

It provides an MCP server implementation of Anthropic's think tool. This exposes dedicated scratchpad functionality to AI models like Claude, allowing them to record structured thought processes, review constraints, and plan multi-step operations before executing actions or returning results.

What tools are provided by MCP Think Tool Server?

The server provides four core tools: think for recording structured thoughts during reasoning tasks, get_thoughts for retrieving previously recorded thoughts, clear_thoughts for resetting stored thoughts between independent sessions, and get_thought_stats for analyzing thinking patterns and thought count metrics.

Which MCP clients work with MCP Think Tool Server?

MCP Think Tool Server works with Claude Desktop and any other client that supports the standard Model Context Protocol over stdio. It can be registered via configuration files such as claude_desktop_config.json using npx or globally installed npm packages.

Is MCP Think Tool Server open source?

Yes, MCP Think Tool Server is an open source project released under the MIT license. Developers can inspect, modify, and contribute to the code via its official GitHub repository at https://github.com/cgize/claude-mcp-think-tool.