The original server's schema had branch parameters with nothing behind them; here creating, switching, inspecting and merging a branch are separate tools, which is what makes 'explore three approaches and compare' a real move rather than a prompt. The persistence is plain SQLite in a file you can open in any viewer — so the reasoning is inspectable outside the assistant that produced it.
A fork of Anthropic's sequential-thinking server that replaced its one tool with 20. Thoughts are numbered and written to SQLite as they are added, branches are a full lifecycle rather than two inert parameters, and a session can be reloaded in a different conversation weeks later.
- Think, revise and complete — add a thought, correct an earlier one with the original kept, close the chain with a conclusion
- Branch, switch_branch, list_branches, get_branch, close_branch and merge_branch, merging as conclusion only, in full, or as a summary
- get_thought and get_history to retrieve one thought by number or the history filtered by branch
- Tag and search — semantic tags such as hypothesis, evidence or decision, then search by content or by tag
- Export as markdown or JSON, and visualize as an ASCII or Mermaid diagram of the thought structure
- session_save, session_load, session_list and session_summary — the last returning a compressed version for cheap context loading
- Reset to clear the workspace and start again
- Write-through persistence: every think, branch, tag and revise call hits disk as it happens
Nothing but npx, or a clone if you prefer. The database lands in the working directory by default; MAXENTIAL_DB_PATH moves it, and setting it to an in-memory value turns persistence off. If SQLite cannot initialise the server falls back to in-memory rather than failing. You do not call these tools yourself — you ask the assistant to think this way and it does the calling.
One command — npx -y @bam-devcrew/maxential-thinking-mcp