tb-query is a CLI tool and MCP server that lets AI agents and developers query and analyze TensorBoard event files without running a TensorBoard server. It reads events.out.tfevents.* files to extract scalar data, filter by tags and steps, compute statistics, and measure correlations between metrics. The MCP interface exposes tools for querying data, listing tags, locating event files, retrieving tag steps, computing tag statistics, and calculating correlations, enabling integration with AI coding assistants. It supports starting the MCP server with tb-query-mcp and the optional TB_QUERY_EVENTS_PATH environment variable to point to a default logs directory; an event_files resource lists available files. The project also provides a Python API (tb_query.core) and a variety of integration examples for Claude Desktop, Cline, Zed, and Continue. Error handling uses ValidationError for file access or parsing issues.