gemini-ops-agent
MukundaKatta/gemini-ops-agent
Production-incident investigation agent on Google Cloud Agent Builder (ADK) + Gemini 2.5 + Dynatrace MCP. Apache-2.0.
Instalação
npx convez-ai MukundaKatta/gemini-ops-agent1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
gemini-ops-agent
A production-incident investigation agent built on Google Cloud Agent Builder (ADK), Gemini 2.5, and the Dynatrace MCP server.
Live demo: https://gemini-ops-agent-1029931682737.us-central1.run.app Demo video: https://storage.googleapis.com/geminilens-demo-mukunda/gemini-ops-agent-demo.mp4 (2:18, 3.3 MB) License: Apache 2.0
What it does
You hand the agent a one-line symptom report ("checkout-api latency just spiked"). It uses the Dynatrace MCP tools to gather evidence, reasons over the metrics and logs with Gemini, and returns a structured root-cause analysis: cited Problem ID, specific numbers, timestamps, and an actionable next step for the on-call.
The agent uses the standard Dynatrace MCP tool surface (list_problems,
execute_dql, find_entity_by_name, generate_dql_from_natural_language,
etc.) — same as the official dynatrace-oss/dynatrace-mcp server. A local
stub MCP server ships with the repo so you can run end-to-end demos
without a Dynatrace tenant; flip one flag and the same agent code targets
a real tenant.
Architecture
┌─────────────┐ user question ┌─────────────────────────────┐
│ Streamlit │ ───────────────────▶ │ ADK LlmAgent (Gemini 2.5) │
│ dashboard │ │ on Vertex AI │
└─────────────┘ ◀── final answer ──── └────┬────────────────────────┘
│ MCPToolset / stdio
▼
┌─────────────────────────┐
│ Dynatrace MCP server │
│ (stub by default, │
│ real tenant via flag) │
└─────────────────────────┘
Try it locally (no Dynatrace tenant needed)
git clone https://github.com/MukundaKatta/gemini-ops-agent
cd gemini-ops-agent
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
# ADC for Vertex AI:
gcloud auth application-default login
export GOOGLE_CLOUD_PROJECT=your-project
export GOOGLE_GENAI_USE_VERTEXAI=true
export GOOGLE_CLOUD_LOCATION=us-central1
# Streamlit dashboard:
PYTHONPATH=src streamlit run app/dashboard.py
Open the sidebar, type a production symptom, click Investigate.
Try it against a real Dynatrace tenant
Set the official server credentials and flip the stub flag:
export DT_ENVIRONMENT_URL=https://abc12345.live.dynatrace.com
export DT_PLATFORM_TOKEN=dt0c01.XXXX...
In the dashboard sidebar, untick "Use stub Dynatrace MCP". The agent now
spawns the official @dynatrace-oss/dynatrace-mcp-server via npx and
talks to your real tenant.
Programmatic use
from gemini_ops_agent.runner import ask
resp = ask("What's wrong with checkout-api?", stub=True, model="gemini-2.5-flash")
print(resp.final_text)
for event in resp.events:
print(event)
Self-observation
Every Gemini call inside the agent loop is wrapped with
GeminiLens so the agent's
own cost, latency, and trace are recorded to
~/.gemini-ops-agent/traces.jsonl. Useful for tracking how much each
investigation costs you in Vertex AI tokens.
Repo layout
src/gemini_ops_agent/
agent.py ADK LlmAgent definition + MCP toolset wiring
runner.py sync Python wrapper around ADK's async Runner
mcp_stub.py stub Dynatrace MCP server (stdio)
app/dashboard.py Streamlit UI
tests/ pytest suite, 12 passing
Dockerfile Cloud Run / Azure Container Apps image
Tests
PYTHONPATH=src pytest -q
License
Apache 2.0. Mukunda Katta, independent developer.