grafana-oncall-mcp

A dedicated Grafana OnCall MCP (Model Context Protocol) server written in Go.

Overview

The upstream mcp-grafana project targets Grafana Cloud and no longer supports grafana-oncall-app, which is archived and deprecated upstream. This MCP server exists to maintain compatibility for self-hosted Grafana OnCall instances that still run grafana-oncall-app or its rebranded successor grafana-irm-app. If you are using Grafana Cloud, use the official mcp-grafana from Grafana instead.

Quick Start (Recommended: uvx)

The fastest way to use this server is via uvx, which downloads a small Python wrapper plus the pre-compiled Go binary from PyPI on first invocation (~50MB total) and caches them. Requires uv.

# Install uv if you don't have it yet
curl -LsSf https://astral.sh/uv/install.sh | sh

# Smoke test
uvx --from grafana-oncall-mcp grafana-oncall-mcp -version
# grafana-oncall-mcp 0.1.0

Prerequisites

• For uvx: uv (Python package manager; handles the Python wrapper + Go binary).
• For Docker: a working Docker daemon.
• For source builds: Go 1.26.x.
• Access to a self-hosted Grafana instance with at least one supported OnCall plugin installed:
  • grafana-oncall-app (legacy, archived upstream) or
  • grafana-irm-app (rebranded successor)
• Grafana service account token with OnCall permissions.
• MCP client capable of stdio, SSE, or Streamable HTTP.

Configuration

Use direct OnCall API mode to call the OnCall Application API directly, bypassing the Grafana plugin resource proxy. This is the recommended mode for self-hosted instances.

export GRAFANA_ONCALL_API_URL="https://oncall-prod-us-central-0.grafana.net/oncall"
export GRAFANA_ONCALL_API_KEY="oncall_user_scoped_key"

Get these values from OnCall -> Settings:

• GRAFANA_ONCALL_API_URL: the OnCall Application endpoint
• GRAFANA_ONCALL_API_KEY: the user-scoped API key

Alternatively, set the Grafana endpoint and service account token:

export GRAFANA_URL="https://example.grafana.net"
export GRAFANA_SERVICE_ACCOUNT_TOKEN="glsa_..."

GRAFANA_API_KEY may be used instead of GRAFANA_SERVICE_ACCOUNT_TOKEN for compatibility with upstream mcp-grafana configuration conventions.

Optional settings:

# Default is oncall-app; accepted values: oncall-app | irm | unset
export GRAFANA_ONCALL_PLUGIN_PREFERENCE="oncall-app"

export GRAFANA_ONCALL_READ_ONLY=true
export OTEL_EXPORTER_OTLP_ENDPOINT="http://localhost:4318"

Build

go mod download
go build -o grafana-oncall-mcp ./cmd/oncall-mcp

Speckit setup (contributors)

This project uses Spec Kit for spec-driven development. Spec Kit manages the feature spec, plan, research, tasks, and contract artifacts under specs/.

1. Install the Specify CLI (requires uv and Python 3.11+):

   uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@v0.10.0
   

2. Initialize Spec Kit in the project root (this project is already initialized; run only if cloning fresh):

   specify init . --force --integration copilot
   

   This creates .specify/memory/constitution.md (gitignored) and registers the Spec Kit slash-commands (/speckit.constitution, /speckit.specify, /speckit.plan, /speckit.tasks, /speckit.implement) in your AI coding agent.

3. For each new feature, run the SDD pipeline from a feature spec:

   /speckit.specify <feature description>
   /speckit.plan <tech stack and architecture choices>
   /speckit.tasks
   

   The generated artifacts (plan.md, research.md, data-model.md, contracts/, tasks.md) are committed alongside the spec in specs/<feature-number>-<name>/.

Run

Stdio (default)

./grafana-oncall-mcp -transport=stdio

SSE

./grafana-oncall-mcp -transport=sse -address=:8000

Connect the MCP client to:

http://localhost:8000/sse

Streamable HTTP

./grafana-oncall-mcp -transport=streamable-http -address=:8000

Connect the MCP client to:

http://localhost:8000/mcp

Tools

Read tools:

• list_oncall_schedules
• get_oncall_shift
• get_current_oncall_users
• list_oncall_teams
• list_oncall_users
• list_alert_groups
• get_alert_group

Write tools (omitted in read-only mode):

• acknowledge_alert_group
• resolve_alert_group
• silence_alert_group
• unresolve_alert_group

RBAC Notes

The token must have permission to:

• Read OnCall schedules, shifts, teams, users, and alert groups
• Write (acknowledge / resolve / silence / unresolve) alert groups

Service account tokens are preferred over API keys. Use read-only mode (GRAFANA_ONCALL_READ_ONLY=true) in environments where the agent must never mutate state.

Running with Docker

A multi-profile docker-compose.yml is provided at the repo root. It supports three transports (stdio / SSE / streamable-HTTP) and reads the same .env file the binary would read directly.

# One-time setup
cp env.example .env
$EDITOR .env                                # fill in GRAFANA_URL + token
docker compose build mcp-stdio              # build the image once

# One-shot stdio (the default profile) — use this from an MCP client:
docker compose run --rm mcp-stdio

# Or run a long-lived HTTP server in the background:
docker compose --profile http up -d
docker compose --profile http logs -f       # check the dual-plugin probe

# Stop the HTTP server
docker compose --profile http down

Coding Agent Setup

The server works with any MCP-aware client (opencode, VS Code Copilot Chat, Claude Desktop, Cursor, Windsurf, etc.). All clients need the same environment variables. The examples below use the recommended direct OnCall API mode (GRAFANA_ONCALL_API_URL + GRAFANA_ONCALL_API_KEY) with GRAFANA_ONCALL_PLUGIN_PREFERENCE="oncall-app" as the default.

Set the env vars in your shell before launching the client:

export GRAFANA_ONCALL_API_URL="https://oncall-<region>.grafana.net/oncall"
export GRAFANA_ONCALL_API_KEY="<OnCall user-scoped API key>"
export GRAFANA_ONCALL_PLUGIN_PREFERENCE="oncall-app"

opencode

Add a server entry to your opencode.json (user-scoped at ~/.config/opencode/opencode.json, or project-scoped at ./opencode.json):

Using uvx (recommended):

{
  "mcp": {
    "grafana-oncall": {
      "type": "local",
      "command": ["uvx", "grafana-oncall-mcp", "-transport=stdio"],
      "environment": {
        "GRAFANA_ONCALL_API_URL": "{{env.GRAFANA_ONCALL_API_URL}}",
        "GRAFANA_ONCALL_API_KEY": "{{env.GRAFANA_ONCALL_API_KEY}}",
        "GRAFANA_ONCALL_PLUGIN_PREFERENCE": "oncall-app"
      },
      "enabled": true
    }
  }
}

Using Docker:

{
  "mcp": {
    "grafana-oncall": {
      "type": "local",
      "command": ["docker", "compose", "-f", "<repo-path>/docker-compose.yml", "--project-directory", "<repo-path>", "run", "--rm", "--no-deps", "--profile", "stdio", "mcp-stdio"],
      "environment": {
        "GRAFANA_ONCALL_API_URL": "{{env.GRAFANA_ONCALL_API_URL}}",
        "GRAFANA_ONCALL_API_KEY": "{{env.GRAFANA_ONCALL_API_KEY}}",
        "GRAFANA_ONCALL_PLUGIN_PREFERENCE": "oncall-app"
      },
      "enabled": true
    }
  }
}

Replace <repo-path> with the absolute path to this repository.

VS Code (GitHub Copilot Chat)

Add to .vscode/mcp.json (workspace-scoped) or VS Code settings.json:

Using uvx (recommended):

{
  "servers": {
    "grafana-oncall": {
      "type": "stdio",
      "command": "uvx",
      "args": ["grafana-oncall-mcp", "-transport=stdio"],
      "env": {
        "GRAFANA_ONCALL_API_URL": "${input:grafanaOncallApiUrl}",
        "GRAFANA_ONCALL_API_KEY": "${input:grafanaOncallApiKey}",
        "GRAFANA_ONCALL_PLUGIN_PREFERENCE": "oncall-app"
      }
    }
  },
  "inputs": [
    {
      "id": "grafanaOncallApiUrl",
      "type": "promptString",
      "description": "OnCall Application endpoint URL",
      "default": "https://oncall-prod-us-central-0.grafana.net/oncall"
    },
    {
      "id": "grafanaOncallApiKey",
      "type": "promptString",
      "description": "OnCall user-scoped API key",
      "password": true
    }
  ]
}

Restart the Copilot Chat session after editing: Command Palette > "Copilot: Restart Chat Session".

Claude Desktop

Add to ~/.claude/claude_desktop_config.json (user-scoped):

Using uvx (recommended):

{
  "mcpServers": {
    "grafana-oncall": {
      "command": "uvx",
      "args": ["grafana-oncall-mcp", "-transport=stdio"],
      "env": {
        "GRAFANA_ONCALL_API_URL": "${GRAFANA_ONCALL_API_URL}",
        "GRAFANA_ONCALL_API_KEY": "${GRAFANA_ONCALL_API_KEY}",
        "GRAFANA_ONCALL_PLUGIN_PREFERENCE": "oncall-app"
      }
    }
  }
}

Using Docker:

{
  "mcpServers": {
    "grafana-oncall": {
      "command": "docker",
      "args": [
        "compose",
        "-f", "<repo-path>/docker-compose.yml",
        "--project-directory", "<repo-path>",
        "run", "--rm", "--no-deps", "--profile", "stdio", "mcp-stdio"
      ],
      "env": {
        "GRAFANA_ONCALL_API_URL": "${GRAFANA_ONCALL_API_URL}",
        "GRAFANA_ONCALL_API_KEY": "${GRAFANA_ONCALL_API_KEY}",
        "GRAFANA_ONCALL_PLUGIN_PREFERENCE": "oncall-app"
      }
    }
  }
}

Restart Claude Desktop after editing the config.

Verifying the connection

After configuring your client, ask the agent:

"List the OnCall teams in my Grafana."

A successful response from list_oncall_teams confirms the connection is healthy.

Release Process (for maintainers)

This repo ships a single Go source tree, but produces two distribution channels from one source:

1. GitHub Releases (.tar.gz / .zip) — built by GoReleaser from .goreleaser.yaml on every v* tag push.
2. PyPI (uvx grafana-oncall-mcp) — built by go-to-wheel which bundles the Go binary into a Python wrapper. Eight platform-specific wheels (linux/darwin/windows × amd64/arm64) are produced and uploaded via OIDC trusted publishing.

The CI wiring lives in .github/workflows/release.yml. The two jobs (goreleaser then pypi) run sequentially on every v* tag push.