Ops Tools MCP Server

One chat window to talk to all your infrastructure — Airflow, EMR, S3, Confluence, and Azure DevOps.

No more jumping between 5 different UIs. Just ask what you want in plain English.

What Is This?

It's an MCP (Model Context Protocol) server that gives AI assistants (like Gemini CLI) access to 44 tools across your entire ops stack. You talk to it in natural language, and it calls the right APIs for you.

Example:

You: "Which DAGs failed today in prod?" AI: calls the Airflow API, gets all runs, filters failures, shows you a summary with diagnosis commands

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Set Up Your .env File

Copy the example and fill in your values:

cp .env.example .env

3. Run the Server

python -m mcp_server.main

The server runs on stdio — connect it to Gemini CLI, VS Code, or any MCP client.

4. Connect from Gemini CLI

Add this to your MCP config (server.json):

{
  "mcpServers": {
    "ops-tools": {
      "command": "python",
      "args": ["-m", "mcp_server.main"],
      "cwd": "D:\\MCP"
    }
  }
}

Environment Setup

This server works across 4 AWS accounts (dev, uat, test, prod). Each is a separate AWS account with its own credentials.

AWS Profiles (via gimme-aws-creds)

AWS_REGION=eu-west-2

AWS_PROFILE_DEV=consumersync-dev
AWS_PROFILE_UAT=consumersync-uat
AWS_PROFILE_TEST=consumersync-test
AWS_PROFILE_PROD=consumersync-prod

MWAA Environments (Airflow)

MWAA_ENV_DEV=eec-aws-uk-ms-dev-consumersyncenv-mwaa
MWAA_ENV_UAT=eec-aws-uk-ms-uat-consumersync-mwaa
MWAA_ENV_TEST=eec-aws-uk-ms-tst-consumersync-mwaa
MWAA_ENV_PROD=eec-aws-uk-ms-prod-consumersync-mwaa

EMR Log Buckets

EMR_LOG_BUCKET_DEV=eec-aws-uk-ms-consumersync-dev-logs-bucket
EMR_LOG_BUCKET_UAT=eec-aws-uk-ms-consumersync-uat-logs-bucket
EMR_LOG_BUCKET_TEST=eec-aws-uk-ms-consumersync-tst-logs-bucket
EMR_LOG_BUCKET_PROD=eec-aws-uk-ms-consumersync-prod-logs-bucket
EMR_LOG_PREFIX=spark-logs

Confluence

CONFLUENCE_BASE_URL=https://pages.experian.local
CONFLUENCE_PAT=your-personal-access-token
CONFLUENCE_SPACE_KEY=ACTIVATE

Azure DevOps (TFS)

AZDO_BASE_URL=https://ukfhpapcvt02.uk.experian.local/tfs/DefaultCollection
AZDO_PAT=your-personal-access-token
AZDO_PROJECT=Activate
AZDO_TEAM=Activate Team

Important: The AI will always ask you "Which environment?" before calling any AWS tool. It never defaults silently — this prevents accidental cross-account mistakes.

All 44 Tools

Airflow / MWAA (11 tools)

Everything you need to monitor, debug, and manage your DAGs.

Common things you'd say:

• "Show me all DAGs in dev"

• "Which DAGs failed today in prod?"

• "How has hem_processing been running lately?"

• "Trigger ttdcustom_processing in uat"

• "Pause digital_taxonomy in prod"

• "Retry the initialise task on yesterday's failed run"

EMR Serverless (10 tools)

Manage Spark jobs, read driver logs, browse S3 log files, track costs.

Common things you'd say:

• "Show me the Spark driver log for this job"

• "What failed in the stdout log?"

• "Cancel that stuck job"

• "Stop that EMR application"

• "Force-stop the app and cancel all running jobs"

• "Delete that EMR application"

• "How much has EMR cost us this week?"

• "Read this S3 file: s3://bucket/path/to/file.csv"

S3 — General (4 tools)

Browse any S3 bucket in the account — not just EMR logs.

Common things you'd say:

• "What S3 buckets do we have in dev?"

• "Show me what's in the raw data bucket"

• "List all CSV files in the raw bucket"

• "How much data is in this S3 folder?"

• "Read this parquet file from S3"

• "How big is this file?"

Confluence (9 tools)

Search, read, and write documentation — without opening a browser.

Common things you'd say:

• "Find documentation about Audience Engine"

• "Read that runbook page"

• "Create a new troubleshooting guide under the runbooks section"

• "What are the child pages under the HEM documentation?"

Pro tip: When you say "docs", "documentation", "wiki", or "runbook", the AI knows to search Confluence automatically.

Azure DevOps / TFS (8 tools)

Sprint tracking, work items, source code — all from chat.

Common things you'd say:

• "What sprint are we in?"

• "What's everyone working on?"

• "Show me PBI 12345"

• "What's in the backlog?"

• "Show me all the files in the hem_processing repo"

• "List all Python files in this repo"

• "What's the folder structure of this repo?"

Orchestration (1 tool)

The power tool — chains multiple tools together for one-shot answers.

What you'd say:

• "Diagnose the failure for hem_processing in prod"

• "What went wrong with ttdcustom_processing yesterday?"

This one tool replaces 5-6 manual steps that used to take 20 minutes.

Utility (1 tool)

How It Works

You (plain English) → AI (Gemini/Claude) → MCP Server → APIs (Airflow, EMR, S3, Confluence, TFS)

1. You type a question in natural language

2. The AI figures out which tool(s) to call

3. The MCP server calls the actual APIs (MWAA, boto3, Confluence REST, Azure DevOps REST)

4. Results come back formatted and readable

5. The AI can chain tools together — e.g. find a failed run → read its logs → show root cause

Architecture

D:\MCP\
├── mcp_server/
│   ├── main.py              # Server entry point + tool registration
│   ├── config.py             # Environment config (4 AWS accounts)
│   └── tools/
│       ├── _aws_helpers.py   # Shared AWS helpers (S3 client, formatting)
│       ├── mwaa_tools.py     # 11 Airflow tools
│       ├── emr_tools.py      # 10 EMR Serverless tools
│       ├── s3_tools.py       # 4 general S3 tools
│       ├── confluence_tools.py # 9 Confluence tools
│       ├── azdo_tools.py     # 8 Azure DevOps tools
│       ├── orchestration_tools.py # 1 orchestration tool
│       └── utility_tools.py  # 1 utility tool
├── .env                      # Your local config (not committed)
├── .env.example              # Template for .env
├── server.json               # MCP client config
├── requirements.txt          # Python dependencies
├── DEMO_SCRIPT.md            # 15-minute demo walkthrough
└── README.md                 # This file

Key Design Decisions

• Fresh credentials every call — No client caching for S3 or EMR. Every API call gets a fresh boto3 session so expired credentials never cause silent failures.

• Environment-aware — All AWS tools require you to specify dev/uat/test/prod. The AI asks if you forget. Each env points to a different AWS account.

• MWAA session cache — The Airflow login token is cached (it needs auth cookies), but the cache clears automatically on 401/403 errors and retries.

• Clean log output — Spark driver logs are auto-decompressed from .gz, Confluence HTML is converted to clean markdown text.

• Interactive responses — DAG runs are numbered so you can say "tell me about run #3". Work items show ready-to-use follow-up commands.

Troubleshooting

Requirements

• Python 3.10+

• VPN access (for MWAA, Confluence, Azure DevOps)

• gimme-aws-creds configured for all 4 AWS accounts

• Confluence PAT

• Azure DevOps PAT

• MCP-compatible client (Gemini CLI, VS Code, Claude Code, etc.)