langfuse-agent¶
Langfuse observability MCP Server + A2A Agent for the agent-utilities ecosystem — typed, deterministic access to the Langfuse tracing, evaluation, prompt, and dataset APIs.
Official documentation
This site is the canonical reference for langfuse-agent, maintained alongside
every release.
Overview¶
langfuse-agent wraps the Langfuse REST surface with typed,
deterministic MCP tools and a Pydantic-AI agent server. It provides:
LangfuseApi— arequests-based REST facade over the Langfuse API, organized by domain (observability, datasets, prompts/models, management, annotation queues).- A current catalog of 5 action-routed and 81 one-to-one tools
(
langfuse-mcpconsole script): traces, observations, scores, sessions, datasets, prompts, models, projects, organizations, SCIM, graph ingestion, and the OpenTelemetry export surface. The default intent mode discloses exact tools only when needed. - An A2A agent server (
langfuse-agentconsole script) that auto-discovers the MCP tools and routes requests through the agent-utilities graph engine.
The connector remains inactive when credential references are absent. GraphOS resolves the references only when it launches the provider.
Explore the documentation¶
- Installation — pip, source, extras, and the prebuilt Docker image.
- Deployment — run the MCP and agent servers, Docker Compose, Caddy + Technitium.
- Usage — the MCP tools, the
LangfuseApiclient, and the CLI. - Backing Platform — deploy Langfuse with Docker.
- Overview — the full tool surface and ecosystem role.
- Concepts — the
CONCEPT:LF-*registry.
Quick start¶
Install the package during provisioning, configure AgentConfig, and start the
already-installed GraphOS runtime:
python -m pip install "langfuse-agent[mcp]"
export LANGFUSE_HOST=https://langfuse.example.invalid
export LANGFUSE_PUBLIC_KEY_REF=env://LANGFUSE_PROJECT_PUBLIC_KEY
export LANGFUSE_SECRET_KEY_REF=env://LANGFUSE_PROJECT_SECRET_KEY
export LANGFUSE_TLS_PROFILE_REF=env://LANGFUSE_RUNTIME_TLS_PROFILE
graph-os
GraphOS registers the provider lazily and starts the installed
langfuse_agent.mcp_server module with its own interpreter. No package download
occurs at runtime. The TLS-profile reference is optional when system trust is
sufficient.
See Installation and Deployment for the full matrix (PyPI extras, Docker image, all transports, the agent server, reverse proxy, DNS).