dragonejt/iceberg-mcp-server: MCP Server for Apache Iceberg
dragonejt’s iceberg-mcp-server gives AI agents the ability to modify Apache Iceberg data, not just read it or run SQL queries.
3 comments
This is cool, it's unfortunate that most open source projects (even large and important ones) don't have an MCP server unless they have a direct corporate backer like Supabase or ClickHouse.
How do you communicate differences in PyIceberg / DuckDB capabilities to the agent, seems like that could get hairy pretty quickly?
iceberg-mcp-server uses PyIceberg to connect to the catalog first, to determine what type of catalog it is. It then uses the catalog type to determine if DuckDB supports connecting to that type of Iceberg catalog at all. PyIceberg supports almost all Iceberg catalog types under the sun, but DuckDB only supports REST including Snowflake Polaris and standard URI/access token, Glue, and S3 Tables catalog types.
I only use DuckDB for a query tool right now for SQL queries, whereas all the other tools for table and namespace interactions are based on PyIceberg.
iceberg-mcp-server is an MCP Server for Apache Iceberg, enabling users to read, query, and manipulate data within Iceberg catalogs. It supports reading and data manipulation using catalog types supported by PyIceberg, and supports SQL queries using catalog types compatible with DuckDB.
I've been working on it occasionally over the past year, and now it has finally gotten to a state where all the basic features are implemented. I'm looking for feedback and ideas to implement, and maybe even fellow contributors.
Also available on PyPI: https://pypi.org/project/iceberg-mcp-server/