The session ID is what separates this from thinking out loud: the history outlives the turn, so a plan begun yesterday is resumed rather than re-derived. Revisions and branches are recorded instead of overwriting what came before, which means the abandoned path is still there when it turns out to have been the right one.
A local server with three tools that record sequential reasoning: write one step, read the history back, and clear it. Steps can revise an earlier one or branch from it, and any tool recommendations attached to a step are validated against what you said was available rather than invented by the server.
- Record one reasoning step at a time, with its number, the total planned, and whether another is needed (sequentialthinking_tools).
- Revise or branch: is_revision and revises_thought correct an earlier step, branch_from_thought and branch_id follow an alternative line without discarding the first.
- Attach available_tools and recommended_tools to a step and have the recommendation validated against the available list.
- Read a session's thoughts back, by branch or limited in number, to inspect or resume it (get_thinking_history).
- Clear one session, or all of them (clear_thinking_history).
Node.js on the machine; no account and no key. MAX_HISTORY_SIZE bounds how much is retained. MIT licensed.
One command — npx -y mcp-sequentialthinking-tools
