Judgeval SDK Reference
Reference documentation for the Judgeval SDK
Use the Judgeval Python or TypeScript SDK to query project data with SQL, trace agent execution, run evaluations, and manage datasets.
For programmatic queries, call sql(...) after inspecting the Python SQL reference or TypeScript SQL reference. You can also fetch the schema with Python's discover_schema() or TypeScript's discoverSchema(). The SQL guide covers setup, query examples, and limits. For agent workflows, use the Judgment MCP server.
The client's api property has a method for each MCP server tool backed by an API endpoint, named after the tool. Each method calls the matching endpoint in the REST API reference. For example, client.api.list_projects() in Python and client.api.listProjects() in TypeScript list your project summaries.
Before using the reference
Install and configure judgeval before constructing a client. Use a personal API key for queries and for dataset or test-suite management. Methods may also need a saved judge, populated dataset or recorded traces. Follow the corresponding task guide to create those resources before copying a reference snippet.
Core SDK Components
Judgeval
Primary client for SQL queries, evaluations, and datasets
API client
Methods named after the MCP server tools, available as client.api
SQL
Discover the virtual schema and run read-only queries over project data
Tracer
Capture and monitor agent execution traces for debugging and analysis
Dataset
Manage collections of examples for batch evaluation
Judge
Create Code judges with typed responses