Overview

The Metoro MCP Server bridges your Metoro observability backend and Claude Desktop via the Model Context Protocol (MCP). Run the server locally and register it as an MCP provider in Claude Desktop so an LLM can query Metoro APIs and act on Kubernetes observability data — metrics, traces, service maps, and resource metadata — without leaving your chat interface.

This makes it easier to perform cluster investigations, answer operational questions, and automate routine tasks from natural-language prompts. Developers and SREs can use Claude to surface telemetry, find high-latency services, inspect recent traces, and drill into pod/state details backed by Metoro’s eBPF-powered data.

Features

Exposes Metoro APIs to any MCP-capable client (e.g., Claude Desktop).
Read-only access to cluster telemetry (metrics, traces, service and pod metadata).
Lightweight Go server built on the Metoro MCP Golang SDK.
Works with your Metoro account or with the public demo cluster for evaluation.
Simple local configuration for rapid setup and integration with Claude Desktop.

Installation / Configuration

Prerequisites:

Go (1.20+ recommended)
Claude Desktop app (to register the MCP server)
Metoro auth token (or the demo token) and API URL

Clone and build the server:

git clone https://github.com/metoro-io/metoro-mcp-server.git
cd metoro-mcp-server
go build -o metoro-mcp-server

Run the server locally:

# run with env vars set inline
METORO_AUTH_TOKEN="your_token" METORO_API_URL="https://us-east.metoro.io" ./metoro-mcp-server

Required environment variables:

Variable	Description
METORO_AUTH_TOKEN	Your Metoro API token (from Metoro Settings → User Settings)
METORO_API_URL	Base URL of your Metoro API (e.g., https://us-east.metoro.io or demo URL)

Registering with Claude Desktop

Install Claude Desktop: https://claude.ai/download
Edit (or create) the Claude Desktop config JSON at:
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Linux/Windows: refer to Claude Desktop documentation for the config path
Add an MCP server entry. Example (macOS path shown):

{
  "mcpServers": {
    "metoro-mcp-server": {
      "command": "/path/to/metoro-mcp-server",
      "args": [],
      "env": {
        "METORO_AUTH_TOKEN": "YOUR_METORO_TOKEN",
        "METORO_API_URL": "https://us-east.metoro.io"
      }
    }
  }
}

Restart Claude Desktop. The new Metoro MCP Server should appear in the MCP servers dropdown.

Tip: If you don’t have a Metoro account, use the demo API URL and demo token included in the repository README. Demo tokens are public and intended for exploration only.

Available Resources

The server surfaces Metoro API resources to the MCP consumer. Typical resource types include:

Clusters and cluster metadata
Namespaces
Services and service dependencies (service map)
Deployments and Pods
Metrics (latency, request rate, error rate)
Traces and span details
Alerts and incidents
Logs/Recent events (if available via Metoro APIs)

These resources are presented as conversational tools the LLM can call to form answers and suggest actions. The exact operations are governed by the server’s MCP handlers and the Metoro API surface.

Use Cases

Quick incident triage:
- Prompt: “Which services in namespace billing have the highest 95th-percentile latency in the last 10 minutes?”
- Outcome: The LLM queries metrics via the MCP server and returns a ranked list with timestamps and links to traces.
Root-cause assistance:
- Prompt: “Show recent traces for service payments that experienced errors and list the top failing endpoints.”
- Outcome: The server fetches traces and the LLM aggregates error-prone endpoints and suggested next steps.
Resource inventory and change impact:
- Prompt: “What deployments changed in the last 24 hours in prod and which pods restarted?”
- Outcome: The LLM reports recent deployment rollouts, restarted pods, and potential correlated alerts.
Onboarding and troubleshooting for junior engineers:
- Prompt: “Explain why request rates to service X dropped and show related logs or events.”
- Outcome: The LLM synthesizes metrics, alerts, and relevant events to produce a readable diagnostic.
Demo and learning:
- Use the demo cluster to explore sample telemetry and practice natural-language queries without a production account.

Troubleshooting & Security Notes

Invalid token: 401 or authentication errors mean the METORO_AUTH_TOKEN is incorrect or expired. Refresh the token in Metoro Settings.
Network/firewall: Ensure the host running the MCP server can reach METORO_API_URL.
Claude integration: Edit the correct claude_desktop_config.json for your OS and restart the app after changes.
Security: The server reads your Metoro token from environment variables. Treat the token like any API secret and store it securely (don’t check it into VCS).

Development & Extensions

The server is implemented in Go and uses the metoro-io/mcp-golang SDK. You can extend handlers to expose additional Metoro endpoints or add write-capable operations (be cautious and add explicit authorization if enabling mutating actions).
Running tests: see repository’s Go test workflow; add unit tests when you extend API handlers.

Resources

Metoro MCP Server repo: https://github.com/metoro-io/metoro-mcp-server
Model Context Protocol: https://modelcontextprotocol.io
Metoro observability: https://metoro.io
MCP Golang SDK: https://github.com/metoro-io/mcp-golang

This setup enables natural-language-driven observability workflows inside Claude Desktop, speeding up investigations and making Metoro telemetry accessible in conversational automation.

Metoro MCP Server: Kubernetes Control via Claude Desktop