# IMDb SQL

> SQL (DuckDB dialect) over IMDb's public dataset, refreshed daily: ~8M rows covering every rated title (movies, series, episodes, shorts, games). The website runs queries in the browser; agents can use the MCP server below.

## MCP server

- URL: https://imdb-sql.fiodorov.es/mcp
- Transport: Streamable HTTP (stateless, JSON responses). No authentication.
- Tools:
  - `query_imdb(sql)`: run one read-only SELECT against `FROM 'imdb.parquet'`; returns `{columns, rows, rowCount, truncated}` (max 500 rows).
  - `get_imdb_schema()`: columns, types, query tips, and which dataset build is live.

Add it to Claude Code: `claude mcp add --transport http imdb-sql https://imdb-sql.fiodorov.es/mcp`. In Claude.ai or ChatGPT, add a custom connector with the URL above.

## Data

One row per title per regional/alternate title. Columns: titleId, title, region (NULL = original-title row), language, primaryTitle, titleType (movie, tvSeries, tvEpisode, ...), startYear, genres (comma-separated), averageRating, numVotes.

- Add `region IS NULL` for ~one row per title.
- Filter on `numVotes` first; rows are sorted by it, so it's the fastest filter.
- "Best of" ranking used by the site: `ORDER BY (numVotes * averageRating + 700000) / (numVotes + 100000) DESC`.

Example:

    SELECT primaryTitle, startYear, averageRating, numVotes
    FROM 'imdb.parquet'
    WHERE region IS NULL AND titleType = 'movie' AND numVotes >= 50000
    ORDER BY (numVotes * averageRating + 700000) / (numVotes + 100000) DESC
    LIMIT 20

## Links

- Website: https://imdb-sql.fiodorov.es/
- Source: https://github.com/afiodorov/imdb-sql