MCP Cloud Operations Agent

Overview

An MCP tool server for cloud operations that exposes read tools only. GET /tools lists list_gke, describe_vm and get_billing, and POST /call runs one of them; any request to apply changes is refused. It shows how an agent can inspect GKE clusters, VMs and billing without being able to change them.

• Built with: Python, pytest, GitHub Actions

• Scope: Local laptop proof with tests; it calls no hosted model and refuses apply.

Related MCP server: Google Ads MCP

Project guide

Project architecture · Interview questions and answers

Use the architecture document for the component diagram, implementation boundaries, and verification entry points. The interview guide includes source-backed answers and project walkthroughs.

Implementation map

Local setup and verification

From the repository root (the commands follow the checked-in manifests):

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python -m pytest -q

To serve the FastAPI application locally, install the server separately if it is not already available:

python -m pip install uvicorn
PYTHONPATH=src python -m uvicorn mcpcloud.main:app --reload

Level: 10 — MCP & Tool-Using Agents

Skills: Python, cloud read tools

Tools: list_gke, describe_vm, get_billing. Apply refused.

pip install -r requirements.txt
pytest -q

This is a local laptop proof. It does not call a hosted model and it does not apply production changes.

Ops plane

Workspaces, tenant isolation, job approval, and audit live under /v1. Production apply is refused. See docs/ARCHITECTURE.md.

Documentation checks

Project architecture, interview guides, and local source links are checked automatically on pushes and pull requests. Run the same check locally:

python3 .github/scripts/validate_project_docs.py

See service improvements and local run instructions.