Summary
A query language for pages, not a scraper.
The interesting move is that the docs are fetchable by the model itself, so the assistant does not need to already know LSD SQL — it reads the language reference, then writes against it.
What it is
An MCP server for LSD, a SQL-like language that treats the web as a Postgres-compatible database. Give it a URL and the assistant writes LSD SQL, self-corrects it, and runs it against LSD's cloud browsers.
What you get
- Structured results out of a page without you writing a scraper for it
- The assistant writing LSD SQL, watching it fail, and correcting it before running again
- Queries executed against cloud browsers rather than on your machine
- The language's own documentation reachable from inside a session — `SCAN https://lsd.so/docs/database/language` returns it as markdown, which is how the model learns the syntax mid-conversation
- Just-in-time tables, so a query returns data without a schema being declared first
Requirements
Python and uv installed, plus an LSD account and an API key from your profile page — free to create with a Google account. Put `LSD_USER` and `LSD_API_KEY` in `.env`, then run `uv run mcp install app.py` and restart your client. If Postgres has never been on the machine you may hit `Error: pg_config executable not found.` — install postgres and rerun.