Eagle-Eye: the 'On-Call & Observability MCP Server'

Eagle-Eye is a set of read-only MCP servers created to assist with on-call duties and provide Observability insights to make on-call life easier using the power of AI.

Eagle-Eye interacts with the following services using the MCP servers:

• Kubernetes (k8s)
• PagerDuty
• Prometheus
• Datadog
• ...many more coming soon! are in the process of adding more built-in MCP servers.

It's written in Python using the FastMCP framework. The MCP server code is located in respective directories under the tools/ folder.

MCP protocol overview

MCP is an open protocol created by Anthropic (founders of Claude AI) that standardizes the way AI apps connect and interact with external tools, services, and data sources using the APIs those services expose, its like a proxy that sits between the AI client and the service.

• https://docs.anthropic.com/en/docs/agents-and-tools/mcp
• https://blog.bytebytego.com/i/159075598/what-is-mcp

Why the need for an in-house MCP server?

• Most third-party MCP servers are open source and built by unknown authors. They pose security risks (e.g., potential backdoors, data exfiltration).
• We currently lack formal guardrails for their use. For example, we could add a third-party MCP server to Cursor IDE/Qcli and request it to fetch pods from Kubernetes—but instead of fetching them, it could delete the pods.
• There's some good reads on MCP security issues:

-- https://invariantlabs.ai/blog/mcp-github-vulnerability

-- https://www.lasso.security/blog/why-mcp-agents-are-the-next-cyber-battleground

Project Setup

This project uses uv for Python environment and package management. Recommended First-Time Setup:

1. Clone the repo and switch to the develop branch (currently, the code is not merged to main since active work is going on in this repo to make it better).

2. We've added a Makefile to make installation easier.

   1. Install uv and Setup Environments:

      Step 1: Install uv

      make install-uv

      After this command completes, ensure uv is in your system's PATH. You might need to source your shell profile (e.g., source ~/.zshrc or source ~/.bashrc) or open a new terminal window for the changes to take effect.

      Step 2: Set up Virtual Environments & Install Dependencies Once uv is installed and accessible in your PATH, run:

      make setup-all

      This will create virtual environments for all MCP server tools and install their dependencies. You can also just run make, which defaults to setup-all.

Cursor IDE Integration (mcp.json)

To use these MCP servers you need access to Cursor IDE, and configure the mcp.json file. To get access to Cursor IDE please reach out to the IT Team.

...why not chatGPT ? openAI desktop app doesnt yet support MCP servers.

1. Open Cursor IDE.
2. Go to Settings > Cursor settings.
3. Click on Add a global MCP.
4. Paste the JSON configuration below, making the necessary updates for PagerDuty and Datadog credentials.

Important: Three changes need to be made in the JSON below:

• Replace REPLACE_DIRECTORY/eagle-eye/... paths with the correct absolute path to the eagle-eye directory on your system (e.g., /Users/your_username/path/to/eagle-eye/...).

• Replace placeholder values for PAGERDUTY_API_KEY, DD_APP_KEY, DD_API_KEY keys/tokens with the read-only credentials

• Replace PAGERDUTY_USER_TOKEN with your own personal token. Since this token varies with permission, the MCP behavior will be different. You can find it in PagerDuty by clicking My Profile --> User Settings --> Create API User Token

{
  "mcpServers": {
    "k8s-readonly-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "REPLACE_DIRECTORY/eagle-eye/tools/k8s",
        "run",
        "main.py"
      ],
      "description": "Runs the Eagle Eye application using uv."
    },
    "pagerduty-readonly-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "REPLACE_DIRECTORY/eagle-eye/tools/pagerduty-mcp-server/src/pagerduty_mcp_server",
        "run",
        "-m",
        "pagerduty_mcp_server.main"
      ],
      "id": "pagerduty-mcp-server",
      "name": "pagerduty-mcp-server",
      "description": "pagerduty-mcp-server",
      "env": {
        "PAGERDUTY_API_KEY": "YOUR_PAGERDUTY_API_KEY_HERE",
        "PAGERDUTY_USER_TOKEN": "YOUR_PERSONAL USER TOKEN HERE"

      }
    },
    "prometheus-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "REPLACE_DIRECTORY/eagle-eye/tools/prometheus/prometheus-mcp-server/src/prometheus_mcp_server",
        "run",
        "main.py"
      ],
      "env": {
        "PROMETHEUS": "<REPLACE_WITH_PROM_URL>"
      }
    },
    "datadog-readonly-mcp-server": {
      "command": "uv",
      "args": [
        "--directory",
        "REPLACE_DIRECTORY/eagle-eye/tools/datadog-mcp-server/src/datadog_mcp_server",
        "run",
        "main.py"
      ],
      "description": "datadog-mcp-server",
      "env": {
        "DD_API_KEY": "YOUR_DATADOG_API_KEY_HERE",
        "DD_APP_KEY": "YOUR_DATADOG_APP_KEY_HERE"
      }
    }
  }
}

Demo

To check why a Kubernetes pod is crashing, you can ask Cursor (make sure to switch to the correct cluster using kubectx), and to investigate a Datadog alert, you can use Eagle-eye.

Roadmap

More integrations and automations to come...

Credits

• https://github.com/Flux159/mcp-server-kubernetes
• https://github.com/wpfleger96/pagerduty-mcp-server
• https://github.com/MaitreyaM/WEB-SCRAPING-MCP