Harbor MCP Server: Let AI Query Container Images and Registries

Why Give AI Access to Harbor?

Harbor is the CNCF-graduated container registry that enterprises use to store, sign, and scan Docker images. When you need to check if an image exists, review vulnerability scan results, or verify image signatures before deploying, you navigate between the Harbor UI and CLI tools. During release processes, those context switches add friction.

An MCP server for Harbor lets your AI agent query images, check scan results, verify signatures, and inspect repository policies directly from Claude, Cursor, or Windsurf.

What Becomes Possible

Once your AI agent has Harbor access:

• "What tags exist for the payments-service image?"
• "Are there any critical vulnerabilities in the latest API image?"
• "Which images were pushed in the last 24 hours?"
• "Is the production tag signed and verified?"
• "How much storage is the staging project using?"
• "What retention policies are set on the release repository?"

How It Works

1. Go to DataFaucet and enter your Harbor URL (e.g. https://harbor.company.com)
2. Log into your Harbor instance
3. Browse projects, repositories, and scan results as you normally would
4. Each action becomes an MCP tool your AI can call
5. Deploy and connect to Claude, Cursor, or Windsurf

Works with Harbor OSS and Enterprise. The AI agent gets whatever access your robot account provides.

Security Best Practices

• Robot accounts with read-only scope — create dedicated robot accounts limited to pull and list
• Project-scoped — restrict to specific projects, not registry-wide
• No push/delete access — prevent AI from modifying images or tags
• Audit logging — Harbor logs all API access with identity attribution
• Token expiration — set robot account tokens to expire regularly

Use Cases by Role

Platform engineers: Registry management. "How many images are in the production project?" without navigating the Harbor UI.

Backend developers: Pre-deploy checks. "Any critical CVEs in the image I just built?" without running trivy manually.

Security engineers: Vulnerability oversight. "Which images have unscanned tags?" for compliance reviews.

SREs: Incident response. "What image tag is running in production and when was it pushed?" for rollback decisions.

Harbor UI vs AI Agent

Getting Started

Create a robot account for AI access in Harbor:

1. Go to Administration → Robot Accounts → New Robot Account
2. Set permissions: Pull and List only (no Push/Delete)
3. Scope to specific projects
4. Set expiration (e.g. 30 days)
5. Copy the token

Then point DataFaucet at your Harbor instance, authenticate with the robot account, and your AI agent can query images and scan results as MCP tools.