Connecting AI Agents to GitLab: How I Implemented MCP Resources in GitLab Core

Description and use

Implemented MCP resources/list and resources/templates/list endpoints in GitLab Core, enabling LLM agents to access repository context. Added enum constraints to the search scope to prevent invalid inputs.

Problem

MCP clients could not access GitLab resources because the server did not advertise or implement the required endpoints, causing 404 errors and disabling context awareness.

Outcome

Implemented MCP resources and templates endpoints in GitLab Core, resolving 404 errors and enabling AI agents to access project context. Added enum constraints to search scope, reducing hallucinations and server errors.

Workflow

• 1. Identify missing MCP endpoints causing 404 errors.
• 2. Update the server handshake to advertise resources capability.
• 3. Implement ResourcesListHandler to return a list of available resources.
• 4. Implement ResourceTemplatesListHandler to return URI templates for resources.
• 5. Add enum constraints to the search scope JSON schema to prevent hallucinations.
• 6. Verify handlers via GitLab CI pipelines (Danger Bot, RuboCop, Predictive RSpec).

Evidence

Because GitLab's server had not implemented or advertised these endpoints, the server responded with: "Method not found: resources/list"

Source

In the server's handshake responder, we registered the resources capability: ...

Source

Next, we added request dispatchers for the resource endpoints: ...

Source

In Merge Request !259119, we introduced explicit enum constraints to the JSON Schema: ...

Source