Labsco
MCP SERVER

Drawdb MCP

by anatoly314

Build and edit database diagrams from a conversation — the backend owns the diagram, browser optional.

Data Platform: Pipelines, Warehousing, BI & Governance
Summary
The diagram lives in the backend, so the agent is not driving a UI.

Most editor integrations break the moment nobody has the tab open. Making the backend authoritative means schema work happens headlessly and the browser is just a window onto it — which is also why exports render correctly with no GUI running. Worth knowing it has diverged from upstream DrawDB on purpose and only cherry-picks fixes.

What it is

A fork of DrawDB with an MCP server in front of it. The backend holds the authoritative diagram state, so an assistant can create tables, change relationships and export SQL whether or not the editor is open; when a browser is connected it acts as a synced view.

What you get
  • `export_sql` — the diagram as SQL DDL for the current database type, covering PostgreSQL, MySQL, SQLite, MariaDB, MSSQL and Oracle
  • `export_dbml` and `import_dbml` — round-trip through Database Markup Language, which is human-readable and database-agnostic
  • `export_diagram` and `import_diagram` — the complete diagram state as JSON, for backup or restore
  • Every tool works with no browser open; a tab that reconnects adopts whatever the backend holds, and edits made in the editor are submitted back
Requirements

Docker is the quickest path: run `ghcr.io/anatoly-lab/drawdb-mcp:latest` publishing ports 8080 and 3000, open the GUI at `http://localhost:8080`, and connect with `claude mcp add --transport http drawdb-mcp http://127.0.0.1:3000`. Pin the full version to run a beta — `latest` only tracks stable. For local development you need Node.js 20.19+ or 22.22+ and pnpm 8.15.0+, then `pnpm install` and `pnpm dev`. After any upgrade do a hard refresh in the browser, or the cached frontend keeps running against the new backend. AGPL-3.0, inherited from upstream DrawDB.