The difference from a generic SQL server is that it answers about the schema rather than through it: which tables reference this one, whether the design has problems, whether anything here looks like sensitive data. The offline snapshot is the underrated feature — serialize the metadata once and the assistant can answer structural questions about a production database it has no credentials for, and start faster while doing it.
The SchemaCrawler AI MCP server, run from a Docker image, which reads database metadata and answers questions about structure, relationships and schema quality.
- The tables in a database, and the columns in any one of them
- Foreign key relationships — which tables reference the one you are looking at
- Design review over the schema, surfacing problems rather than just listing objects
- Sensitive data spotted from the schema metadata
- SQL written against the real structure, so joins use the keys that exist
- Offline operation from a serialized snapshot, when the database is not reachable — and faster startup even when it is
- Two companion agents in the repository, database-expert and sql-query-assistant, for the two tasks people bring most often
Docker with Docker Compose. Your database connection goes in the Compose file or, on the Docker MCP Catalog path, in Docker Desktop's settings for the server. Health is reachable at the server's /health endpoint on port 9292 once it is up. For offline use, generate the snapshot with SchemaCrawler's serialize command at maximum info level, then point SCHCRWLR_SERVER at offline and SCHCRWLR_OFFLINE_DATABASE at the file.
One command — docker pull mcp/schemacrawler-ai