AyushRaj005/SRE-Debugging-and-Automation-MCP-Server

SRE Debugging & Automation using MCP Servers

A Practical Implementation of AI-Assisted Observability, Debugging, and Workflow Automation Developed and Research during my internship

Overview

This repository contains two example Model Context Protocol (MCP) servers I built as part of my internship work focused on SRE (Site Reliability Engineering), observability, and AI-powered debugging. Due to confidentiality, I cannot share the full internal codebase, infrastructure, or production-grade MCP servers built for Sherlocks.AI. However, this repository showcases:

• The architecture,
• The design patterns,
• The concepts,
• And two simplified MCP server examples

…that represent the type of functionality I engineered during my internship.

My actual internship work included advanced integrations with:

• Sentry
• Datadog
• New Relic
• Coralogix
• AWS EKS / ECS / EC2
• Internal engineering APIs
• Custom incident-automation pipelines

What I Worked On During the Internship

During the internship, I developed multiple MCP servers enabling Claude AI to directly interact with:

Cloud Infrastructure

• Query EKS/ECS cluster health
• Analyze pod crashes, deployments, scaling failures
• Recommend remediation steps

Observability Platforms

• Fetch and interpret real-time KPIs (latency, CPU, 95th percentile, throughput)
• Correlate APM traces (New Relic, Datadog)
• Retrieve logs, errors, alerts, incident states

Debugging Automation

• Autonomous error triage
• Root-cause summarization using LLMs
• Incident timeline reconstruction
• Automated ticket updates and reporting

Developer Experience Tools

• Build MCP command interfaces
• Create endpoint schemas for debugging
• Integrate Claude Desktop with MCP servers
• Optimize development workflows using AI

All production versions are private, but the concepts are recreated here.

What This Repository Contains

This repo includes two simplified MCP servers illustrating the same concepts I used in production:

1. Example JSON MCP Server

A lightweight MCP server that serves JSON resources with schemas.

Features:

• Demonstrates MCP tool definitions
• Shows server registration flow
• Teaches how JSON resources are exposed to Claude
• Useful for configuration debugging

Code: /json-server/

2. Example Directory MCP Server

A filesystem-based MCP server enabling:

• File search
• File preview
• Directory traversal
• Secure file queries

This mirrors the file-system-observability logic I built internally.

Code: /directory-server/

Architecture I Worked With (High-Level)

Here is a simplified architecture diagram of the debugging pipeline:

Claude Desktop
      |
      ↓
MCP Server (FastAPI / Node)
      |
      ↓
Platform Integrations
(AWS / Datadog / Sentry / NR / Internal APIs)
      |
      ↓
LLM-Powered Debugging
- Incident summaries
- RCA explanations
- Log + metric correlation
- Deployment analysis

My work involved building these MCP layers so Claude could perform tasks like:

“Show me all pods crashing in production in the last 15 min.” “Correlate Sentry errors with Datadog CPU spikes.” “Explain this stack trace and suggest fixes.” “Why is my EKS node not scaling?” “Check the 95th percentile latency for today.”

Why MCP is a Game Changer for SRE

Model Context Protocol (MCP) turns Claude into an SRE assistant capable of interacting with real infrastructure.

With MCP, Claude can:

• Query observability APIs
• Access server state
• Troubleshoot deployments
• Analyze logs, traces, errors
• Suggest remediation steps
• Correlate multi-source data

It removes the need to switch between 10 dashboards— your debugging becomes conversational.

Example Use Cases (from internship work)

Real Incident Scenario

“Users are reporting 500 errors. Can you debug?”

Claude (via MCP):

1. Checks EKS pod health

2. Fetches Sentry error spike “DB Connection Timeout”

3. Queries Datadog CPU saturation

4. Correlates deployment history

5. Generates RCA

6. Suggests:

   • Increase DB connection pool
   • Restart affected services
   • Improve readiness probes

Installation & Usage

Install MCP CLI

pip install mcp[cli]

Run an example server

mcp run json-server

Connect to Claude Desktop

Add configuration in:

Windows:

%APPDATA%\Claude\claude_desktop_config.json

macOS:

~/Library/Application Support/Claude/claude_desktop_config.json

Example:

{
  "mcp_servers": {
    "json": {
      "command": "python",
      "args": ["json-server/server.py"]
    }
  }
}

🛠️ Tech Stack Used in Real Internship

While repo shows simplified code, actual work involved:

• Python / FastAPI for MCP servers
• Node.js for real-time observability tools
• Datadog API, Sentry API, NR API, Coralogix API
• AWS EKS/ECS/EC2 SDKs
• Claude SDK (prompt engineering + tool schemas)
• Structured logging + monitoring
• Async queues + rate limit handling