GitLab DevOps Agent

An autonomous AI-powered DevOps assistant built with Google ADK 2.0 and Gemini 2.5 Flash. It connects directly to your GitLab instance and performs real engineering tasks: reviewing code, triaging issues, diagnosing pipeline failures, and more. All through natural language.

---

Live Demo

Try the agent live — no setup required:

gitlab-devops-agent-792314829007.us-central1.run.app

Deployed on Google Cloud Run. Point it at any public GitLab project and start asking questions!

---

Features

Capability	Description
List Issues	Fetch all open issues with age, inferred priority, and stale flags
Triage Issue	Analyze a single issue and suggest labels and priority based on content
Auto-Triage All Issues	Bulk-scan every open issue, apply inferred labels, and post a triage comment
Add Issue Comment	Post a comment on any issue programmatically
Close Issue	Close an open issue by state transition
Review Merge Request	Summarize an MR's changed files, reviewers, approvals, and description
Merge Merge Request	Merge an open MR after verifying it is in a mergeable state
Diagnose Pipeline	Fetch failed job logs from a CI/CD pipeline and surface the root cause
Generate Release Notes	Produce categorized markdown release notes from merged MRs over a date range

The agent also connects to GitLab's official MCP server (https://gitlab.com/-/llm/mcp), giving it access to GitLab's full native tool surface alongside the custom tools above.

---

Tech Stack

• Google Agent Development Kit (ADK) 2.0
• Gemini 2.5 Flash via Vertex AI
• GitLab MCP Server — GitLab's official Model Context Protocol server
• Python 3.11+
• python-gitlab — GitLab REST API client
• python-dotenv — environment variable management
• Docker + Google Cloud Run — containerised deployment

---

Setup

1. Clone the repository

git clone https://github.com/your-username/gitlab-devops-agent.git
cd gitlab-devops-agent

2. Create and activate a virtual environment

python -m venv venv

# Windows
venv\Scripts\activate

# macOS / Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Create a .env file inside the gitlab_agent/ directory:

GITLAB_TOKEN=your_gitlab_personal_access_token
GITLAB_URL=https://gitlab.com
GOOGLE_CLOUD_PROJECT=your_google_cloud_project_id
GOOGLE_GENAI_USE_VERTEXAI=1

Variable	Description
GITLAB_TOKEN	GitLab Personal Access Token with api scope — also used to authenticate the MCP server
GITLAB_URL	Your GitLab instance URL (use https://gitlab.com for GitLab.com)
GOOGLE_CLOUD_PROJECT	GCP project ID for Vertex AI access
GOOGLE_GENAI_USE_VERTEXAI	Set to 1 for Vertex AI, 0 for Gemini API (AI Studio)

Never commit your .env file. It is already listed in .gitignore.

To create a GitLab Personal Access Token: GitLab → User Settings → Access Tokens → Add new token with the api scope.

---

Running the Agent

Locally

adk web

Open http://localhost:8000 and select gitlab_devops_agent.

Docker

docker build -t gitlab-devops-agent .
docker run -p 8080:8080 \
  -e GITLAB_TOKEN=your_token \
  -e GITLAB_URL=https://gitlab.com \
  -e GOOGLE_CLOUD_PROJECT=your_project_id \
  -e GOOGLE_GENAI_USE_VERTEXAI=1 \
  gitlab-devops-agent

Open http://localhost:8080.

Google Cloud Run

gcloud run deploy gitlab-devops-agent \
  --source . \
  --region us-central1 \
  --port 8080 \
  --set-env-vars GITLAB_URL=https://gitlab.com,GOOGLE_GENAI_USE_VERTEXAI=1 \
  --set-secrets GITLAB_TOKEN=gitlab-token:latest,GOOGLE_CLOUD_PROJECT=gcp-project:latest

Environment variables set by Cloud Run take precedence over any local .env file.

---

Example Prompts

List all open issues in project my-org/my-repo

Triage issue #42 in project my-org/my-repo

Auto-triage all open issues in project my-org/my-repo

Review merge request !17 in project my-org/my-repo

Diagnose pipeline 998 in project my-org/my-repo

Generate release notes for my-org/my-repo since 2025-01-01

---

Project Structure

gitlab-devops-agent/
├── gitlab_agent/
│   ├── agent.py       # ADK agent definition, tool registration, and MCP server config
│   ├── tools.py       # GitLab API tool implementations
│   └── .env           # Local environment variables (not committed)
├── Dockerfile
├── .dockerignore
└── README.md

---

Security

• Store credentials exclusively in .env locally, or as Cloud Run secrets in production — never hard-code tokens.
• Grant the GitLab token the minimum required scope (api for full access, or read_api for read-only use cases). The same token authenticates both the python-gitlab tools and the GitLab MCP server.
• Treat auto_triage_all_issues as a write operation — it modifies labels and posts a comment on every open issue in the target project.