Zum Inhalt springen
Kitbay
Einreichen
MCP-Server

datakeel

Läuft lokalApache-2.0Aktualisiert am 14. Nov. 2026138GitHub-SterneKein API-Key nötig

Datakeel is a local-first workbench for data and AI datasets that keeps an audit trail: it supports review of multimodal evidence, answers metric questions with traceable SQL, and prepares agent actions that a person must approve.

Was er kann

Datakeel is a local-first workbench for data and AI datasets that keeps an audit trail: it supports review of multimodal evidence, answers metric questions with traceable SQL, and prepares agent actions that a person must approve. Its MCP server, started with uv, runs over stdio or as a loopback-only SSE server and, by the README's count, offers 102 tools: 37 that read and 65 that prepare plans. Approving or running a plan through MCP is refused and logged; that can only happen in the web interface or the CLI. The MCP setup needs only a data directory path. Language model keys are optional, and without them an offline demo provider is used.

Funktionen

  • Answer metric questions with traceable SQL
  • Review multimodal dataset evidence
  • Prepare change plans that a person approves outside MCP