grafana-mcp

christian-schlichtherle/grafana-mcp

3.3

If you are the rightful owner of grafana-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

This repository provides an MCP (Model Context Protocol) server for the comprehensive discovery, editing, and testing of dashboards and folders in one or more Grafana clusters.

Grafana MCP

This repository provides an MCP (Model Context Protocol) server for the comprehensive discovery, editing, and testing of dashboards and folders in one or more Grafana clusters. Its security model is based on labels and folders.

Quick Start

# Install and run with zero configuration (uses localhost:3000 by default)
uv sync
python main.py

# That's it! The server is ready to use with your local Grafana instance.

Features

  • Dashboard Protection: When creating Grafana dashboards, the MCP server labels them to mark itself as the origin
  • Selective Modification: Only dashboards with required protection labels can be updated or deleted
  • Unrestricted Reading: No restrictions for reading dashboards, allowing enhancement of existing dashboards by copying them
  • Folder Restrictions: Configurable access control to specific folders and subfolders
  • Multi-Cluster Support: Manage dashboards across multiple Grafana instances
  • Comprehensive Search: Full support for Grafana's search API with filtering options

Available MCP Tools

Discovery

Discover and explore Grafana resources across clusters.

Cluster Management
  • list_clusters() - List all configured Grafana cluster names
  • check_cluster_health(cluster: str) - Check health and connectivity of a Grafana cluster
Search Operations
  • search(cluster: str, *, query: str = "", tags: list = [], starred: bool = False, folder_uids: list = [], dashboard_uids: list = [], dashboard_ids: list = [], type: str = "", limit: int = 1000, page: int = 1) - Search dashboards and folders with comprehensive filtering options
Dashboard Discovery
  • read_dashboard(cluster: str, dashboard_uid: str) - Read any dashboard configuration and metadata (no restrictions)
  • inspect_dashboard(cluster: str, dashboard_uid: str) - Detailed structural analysis of dashboard panels, variables, datasources, and layout
Folder Discovery
  • list_folders(cluster: str, *, parent_uid: str = "") - List folders, optionally under a parent folder
  • get_folder(cluster: str, folder_uid: str) - Get detailed folder information including hierarchy and permissions
Datasource Discovery
  • list_datasources(cluster: str) - List all datasources in cluster with name, type, UID, and metadata

Editing

Create, modify, and manage Grafana resources with security controls.

Dashboard Management
  • create_dashboard(cluster: str, dashboard_json: dict, *, folder_uid: str = "") - Create new dashboard with automatic protection labels
  • update_dashboard(cluster: str, dashboard_uid: str, dashboard_json: dict) - Update existing dashboard (requires protection labels)
  • delete_dashboard(cluster: str, dashboard_uid: str) - Delete dashboard (requires protection labels)
  • copy_dashboard(source_cluster: str, source_uid: str, new_title: str, *, target_cluster: str = "", folder_uid: str = "", target_uid: str = "") - Copy dashboard with intelligent UID handling and auto-overwrite capabilities
Folder Management
  • create_folder(cluster: str, title: str, *, parent_uid: str = "") - Create new folder, optionally as subfolder
  • update_folder(cluster: str, folder_uid: str, title: str, *, parent_uid: str = "") - Update/rename/move folder to different parent
  • delete_folder(cluster: str, folder_uid: str, *, force_delete_rules: bool = False) - Delete folder with optional alert rule cleanup

Testing

Validate, test, and analyze dashboard quality and data accuracy.

Validation & Quality Assurance
  • validate_dashboard(cluster: str, dashboard_uid: str) - Comprehensive schema validation and best practices checking
  • compare_dashboards(cluster: str, dashboard_uid_a: str, dashboard_uid_b: str, *, compare_cluster_b: str = "") - Compare two dashboards showing structural and configuration differences
Data Verification & Testing
  • snapshot_dashboard(cluster: str, dashboard_uid: str, *, snapshot_name: str = "", expires_hours: int = 24, time_from: str = "now-6h", time_to: str = "now") - Create snapshot with current data for inspection and testing
  • test_panel_render(cluster: str, dashboard_uid: str, panel_id: int, *, width: int = 1000, height: int = 500, time_from: str = "now-6h", time_to: str = "now", save_to_file: str = "") - Render panel as PNG for visual validation

Configuration

Environment Variables

Configure the MCP server using these environment variables:

# Cluster URLs (space-separated key=value pairs, defaults to "localhost=http://localhost:3000")
export GRAFANA_CLUSTERS="localhost=http://localhost:3000 dev=https://grafana.dev.example.com prod=https://grafana.prod.example.com"

# Cluster API tokens (space-separated key=value pairs, optional)
# Only required for clusters that need authentication
export GRAFANA_TOKENS="dev=glsa_yyy prod=glsa_zzz"  # Note: localhost has no token (unauthenticated)

# Protection labels (space-separated, defaults to "MCP")
export GRAFANA_LABELS="MCP Claude"

# Folder restriction (defaults to "/" for root access)
export GRAFANA_FOLDER=/MCP

Local Development Without Authentication

For local development, you can run Grafana in a Docker container without authentication:

# Run Grafana without authentication
docker run -d \
  -p 3000:3000 \
  -e "GF_AUTH_ANONYMOUS_ENABLED=true" \
  -e "GF_AUTH_ANONYMOUS_ORG_ROLE=Admin" \
  -e "GF_AUTH_BASIC_ENABLED=false" \
  --name grafana \
  grafana/grafana:latest

# Configure MCP server for unauthenticated local access
export GRAFANA_CLUSTERS="localhost=http://localhost:3000"
# No GRAFANA_TOKENS needed!

# Or use the default configuration (no environment variables needed!)
# This defaults to localhost=http://localhost:3000 without authentication

# Run the MCP server
python main.py
Mixed Authentication Environments

You can mix authenticated and unauthenticated clusters:

# Local without auth, production with auth
export GRAFANA_CLUSTERS="localhost=http://localhost:3000 prod=https://grafana.prod.com"
export GRAFANA_TOKENS="prod=glsa_abc123"  # Only prod needs a token

Authentication Setup (For Production Clusters)

For production or authenticated Grafana clusters, you need to create service account tokens. Service accounts are the recommended way to authenticate with Grafana's HTTP API (API keys are deprecated).

Note: Skip this section if you're only using local unauthenticated Grafana instances.

Creating Service Account Tokens via Grafana UI

  1. Create Service Account:

    • Navigate to Administration → Users and access → Service Accounts
    • Click "Add service account"
    • Enter a Display name (e.g., "MCP Server")
    • Select appropriate Role: Viewer, Editor, or Admin
    • Click "Create"

  2. Generate Token:

    • Select your newly created service account
    • Click "Add service account token"
    • Enter a Display name for the token
    • Set Expiration date (or select "No expiration")
    • Click "Generate token"
    • Important: Copy and save the token immediately - it's only shown once!
Creating Service Account Tokens via API
# Create service account
curl -X POST -H "Content-Type: application/json" \
  -d '{"name": "mcp-server", "role": "Editor"}' \
  http://admin:admin@localhost:3000/api/serviceaccounts

# Create token for the service account (use ID from previous response)
curl -X POST -H "Content-Type: application/json" \
  -d '{"name": "mcp-token"}' \
  http://admin:admin@localhost:3000/api/serviceaccounts/{id}/tokens
Using Tokens

Tokens should be prefixed with glsa_ and used in the GRAFANA_TOKENS environment variable:

export GRAFANA_TOKENS="prod=glsa_abc123def456 dev=glsa_xyz789uvw012"
Authentication in Requests

The MCP server automatically adds the Bearer token to all requests:

Authorization: Bearer glsa_abc123def456

For more information, see the official Grafana documentation on service accounts.

Installation & Usage

  1. Install dependencies:

    uv sync
# or
pip install -e .

  2. Set environment variables (see Configuration section above)

  3. Run the MCP server:

    python main.py

The server will validate your configuration and start the MCP server, making all tools available to MCP clients.

Example Usage

# List all configured clusters
clusters = list_clusters()
# Returns: ["localhost", "prod"]

# Explore folder hierarchy
root_folders = list_folders("prod")  # List root-level folders
subfolders = list_folders("prod", parent_uid="folder123")  # List subfolders

# Search for dashboards and folders
dashboards = search("prod", query="monitoring", tags=["production"])
folders_only = search("prod", type="dash-folder", query="team")
dashboards_only = search("prod", type="dash-db", starred=True)

# Create and manage folders
folder = create_folder("prod", "Team Alpha", parent_uid="parent123")
updated = update_folder("prod", "folder123", "New Name", parent_uid="new_parent")

# Copy a dashboard into a specific folder
new_dash = copy_dashboard("prod", "abc123def", "My New Dashboard", folder_uid="folder123")

# Create a dashboard with protection labels automatically added
result = create_dashboard("prod", {
    "title": "Test Dashboard",
    "panels": [...],
    "tags": ["custom"]  # MCP labels will be automatically added
}, folder_uid="folder123")

Security Model

  • Label-Based Protection: Dashboards created by this MCP server are automatically tagged with protection labels
  • Selective Operations: Only dashboards with required labels can be updated or deleted
  • Folder Restrictions: Operations can be restricted to specific folder hierarchies
  • Token-Based Auth: Uses secure service account tokens instead of deprecated API keys
  • No Overwrites: Dashboard creation prevents accidental overwrites of existing dashboards