Langfuse 4.70 adds batch evaluation through MCP

Langfuse published v4.70.0 on October 24, 2026, following v4.69.0 on October 23.

Evaluation and Topics access

Version 4.70 adds a runBatchEvaluation MCP tool and includes batch_action_id in batch-run outputs. It also adds read-only public API endpoints for Topics. The release explicitly calls those endpoints unstable; consumers should not assume a frozen interface.

The command palette gains navigation to prompts, datasets, queues and evaluators. Gateway-key pages also gain clearer descriptions and invitations.

Related changes in 4.69

The preceding release adds MCP operations for creating skill versions and managing labels and tags. It also introduces guided AI Gateway setup and per-tool setup instructions after a gateway key is created. Self-hosted gateway visibility is controlled by deployment configuration rather than implying the gateway is always present.

Version 4.70 additionally requires secrets to be supplied again when a webhook or PostHog destination changes. Teams using those integrations should review the change alongside their upgrade process.

These notes describe new access paths and workflow controls, not an evaluation-quality benchmark. See the Langfuse hub.