← Back to changelog

November 11, 2026

Set the default evaluation model and run batch evaluations via MCP

Tobias Wochinger

AI agents can now read, set, and delete a project's default evaluation model and run evaluators over existing observations through the Langfuse MCP server.

The Langfuse MCP server now lets AI agents manage the project's default evaluation model and run batch evaluations on existing data. For example, an agent can switch the project's default evaluation model to a new judge model, then re-score last week's production observations with the new model.

New MCP tools

Get started