Sequential Thinking

You can use the Sequential Thinking MCP Server to guide AI agents through structured, step-by-step reasoning. It breaks complex problems into numbered thoughts, allows revising earlier steps, and supports exploring multiple reasoning paths across isolated sessions for multiple users. You can run it locally, expose it over HTTP for remote clients, or deploy it as a serverless Cloudflare Worker for global access with persistent state.

How to use

To use the Sequential Thinking MCP Server with an MCP-compatible client, start a server instance and select a deployment mode that fits your needs. You can run in local stdio mode for desktop clients, expose an HTTP endpoint for remote access, or deploy to a serverless environment with durable state. Once running, your client sends thoughts in sequence, with the ability to revise or branch paths and to retrieve the full thought sequence at any time. Each active user gets an isolated session, so concurrent work remains independent.

How to install

Prerequisites: ensure you have Node.js installed on your machine. You can verify by running node -v and npm -v in your terminal.

Install dependencies and build the project.

npm install
npm run build

Additional deployment options and usage patterns

Choose the deployment option that best suits your environment.

Option 1: Stdio Mode (local desktop clients, e.g., Claude Desktop). Start the server with the standard startup command.

Option 2: HTTP Server Mode. Expose a REST-like endpoint for remote clients. Start the HTTP server with the designated start command and provide a port and allowed origins as needed.

Option 3: Cloudflare Workers. Deploy serverless with Durable Objects for global distribution and persistent state.

Configuration and endpoints

The server supports multiple deployment methods with corresponding endpoints and configuration options. In HTTP mode, you typically access a Streamable HTTP endpoint for MCP requests. In Cloudflare Workers mode, you interact with REST-like endpoints such as think, sequence, and reset, mapped to the Durable Objects that hold session state.

Security and session management

Sessions are cryptographically secured and isolated per user. The server supports configurable CORS and rate limiting to protect against abuse. Automatic session cleanup ensures sessions that are idle for a defined period are removed to prevent memory leaks.

Troubleshooting tips

If you encounter issues, check that each deployment mode is started with the correct command for that mode, verify that environment variables are set as needed, and confirm that sessions are being created and scoped per user. Use the health or status endpoints to verify the server is running and sessions are being allocated properly.

Examples of typical usage

Think through a problem step by step by adding sequential thoughts, revising as needed, and optionally creating branches for alternative approaches. Retrieve the entire sequence or reset the session to start fresh.

Notes on deployment options

• Stdio mode is suitable for local development and desktop MCP clients. Use the npm start or npx sequential-thinking-mvp-server commands to run. - HTTP mode exposes an endpoint for remote clients; configure the port and CORS origins via environment variables. - Cloudflare Workers provide a serverless path with global distribution and durable state.

License

MIT license.

Tools and endpoints overview

Core tool: sequential_thinking manages the step-by-step thoughts and supports revision, branching, and sequencing. Other tools include get_thought_sequence, get_thought_branch, reset_thinking_session, and get_session_summary to manage and inspect the thinking process.

Available tools

sequential_thinking

Core tool for adding and managing sequential thoughts with numbering, revision, and branching capabilities.

get_thought_sequence

Retrieve the complete sequence of thoughts for the current session.

get_thought_branch

Retrieve a specific branch of alternative reasoning.

reset_thinking_session

Start a new thinking session and clear the current sequence.

get_session_summary

Get a summary of the current thinking session.