Sequential Thinking MCP: what the one tool does and how to add it
The official Sequential Thinking MCP server gives an agent a scratchpad for step by step reasoning it can revise and branch. Setup for Cursor and Claude, and what I saw testing version 2.1.1.
Sequential Thinking is one of the reference servers from the Model Context Protocol project and it is easy to misunderstand. It does not make the model smarter and it does not call another model. It gives the agent a single tool for writing down its reasoning one step at a time, with room to go back and revise a step or try a different branch.
Version 2.1.1 shipped in the v2.3.0 release of the reference servers on November 4. I ran it locally before writing this. Everything below comes from that run and the official README.
What the tool actually does
The server exposes one tool, sequentialthinking. Each call carries one thought plus a little bookkeeping: which thought number this is, how many the agent expects in total, and whether another thought is needed. Optional fields mark a thought as a revision of an earlier one, or start a branch from an earlier thought with a branch name.
The server keeps the history for the connection and answers with where things stand: the thought number, the total, the list of branches and how long the history is. The thinking itself is the model's. The tool is structure around it, which tends to help on tasks where the first plan is likely to be wrong.
