TKCat 1.3.0 adds a read-only MCP server that lets AI assistants query modeled databases. The server comes in a local stdio form (chTKCat_mcp_server) and a remote Streamable HTTP form (chTKCat_mcp_http_server) that runs next to a ClickHouse-backed chTKCat instance. Tools include list_mdbs, search_tkcat, describe_mdb, describe_tables, preview_table and query_mdb for running ClickHouse SQL. Clients like Claude Code, Claude Desktop and Posit Assistant can discover and invoke these tools using TKCat's documented data models, foreign keys and collections to answer questions spanning multiple databases.
Table of contents
1 Documentation was always the point2 What is MCP3 What’s new in TKCat 1.3.04 Why documented databases make a difference5 Built for shared, governed environments6 Getting startedQuestions this post answers
What is the MCP server in TKCat 1.3.0 and what tools does it expose?
TKCat 1.3.0 ships a read-only MCP server that lets AI assistants like Claude Code, Claude Desktop, or Posit Assistant explore and query modeled databases (MDBs) in a chTKCat (TKCat on ClickHouse) instance. It exposes tools such as list_mdbs, search_tkcat, describe_mdb, describe_tables, list_collections, preview_table, and query_mdb for running ClickHouse SQL. Developers wiring AI assistants into existing databases can track MCP server releases like this one on daily.dev.
What is the difference between chTKCat_mcp_server() and chTKCat_mcp_http_server() in TKCat?
chTKCat_mcp_server() runs a local MCP server started by the client on the user's own computer using stdio transport, requiring R and TKCat installed locally. chTKCat_mcp_http_server() runs as a remote service alongside the chTKCat instance using Streamable HTTP transport, so users don't need R or TKCat on their own machine. Teams deciding between local and remote MCP deployments can follow database tooling updates like this on daily.dev.
Share this post