If you need Claude to reason through complex problems with multiple agents and want a full audit trail, this is your server. It exposes two MCP tools using the Code Mode pattern: thoughtbox_search to query available operations and thoughtbox_execute to run JavaScript against a unified SDK. Inside that SDK you get 28 operations for multi-agent workflows (claim problems, propose solutions, peer review, reach consensus), structured thinking patterns (forward, backward, branching, revision), and a persistent knowledge graph. Every thought gets numbered, timestamped, and stored in a reasoning ledger you can visualize in the built-in Observatory UI at localhost:1729. Runs entirely local in Docker, data stays in ~/.thoughtbox. Built for scenarios where you need to see how conclusions were reached.
claude mcp add --transport stdio kastalien-research-thoughtbox uvx thoughtbox