lgtm-mcp¶

LGTM observability API + MCP server for the agent-utilities ecosystem — typed, deterministic tools over Grafana and Prometheus Alertmanager for agentic alerting, silencing, and dashboard operations.

Official documentation

This site is the canonical reference for lgtm-mcp, maintained alongside every release.

Overview¶

lgtm-mcp wraps the Grafana and Prometheus Alertmanager REST surfaces with typed, deterministic MCP tools, so an agent can read firing alerts, manage silences, and operate dashboards without handling HTTP transport directly. It provides:

Api — a dynamic facade (lgtm_mcp.api_client.Api) composed from the Alertmanager and Grafana clients, with credential authentication and SSL handling.
Action-dispatch MCP tools — lgtm_mcp_alertmanager and lgtm_mcp_grafana, each routing a named action to the underlying client method.
An optional agent server — a Pydantic-AI agent (lgtm-agent) that connects to the MCP tool surface for conversational observability operations.

The server remains inactive when credentials are absent, degrading safely rather than raising on an unreachable backend.

Explore the documentation¶

Installation — pip, source, extras, and the prebuilt Docker image.
Deployment — run the MCP server, the agent server, Docker Compose, Caddy + Technitium.
Usage — the MCP tools and the Api Python client.
Backing Platform — deploy the LGTM observability stack with Docker.
Overview — architecture and the dynamic facade.
Concepts — the CONCEPT:LGTM-* registry.

Quick start¶

pip install "lgtm-mcp[mcp]"
lgtm-mcp                         # stdio MCP server (default transport)

Connect it to a Grafana / Alertmanager deployment:

export GRAFANA_URL=http://your-grafana:3000
export ALERTMANAGER_URL=http://your-alertmanager:9093
export LGTM_TOKEN=your_grafana_api_token
lgtm-mcp --transport streamable-http --host 0.0.0.0 --port 8000

See Installation and Deployment for the full matrix (PyPI extras, Docker image, all transports, reverse proxy, DNS).