GitLab MCP Server is a Model Context Protocol server that enables AI assistants to interact with GitLab through natural language. Ask your AI to create issues, review merge requests, analyze pipelines, and much more — all without leaving your editor.
Instead of switching between your editor and GitLab’s web UI, just ask:
Show me all open merge requests in my-project that need review
Why did the pipeline fail on branch feature/auth? Summarize the error and suggest a fix
Create an issue titled "Refactor auth module" with priority label and assign it to me
The server translates these requests into GitLab API calls, executes them, and returns structured results your AI assistant can understand and act upon.
Default 2-tool low-token mode: find canonical actions with exact schemas, then execute canonical GitLab actions
Meta-Tools
Optional domain mode with 33 base tools, 49 on self-managed Enterprise/Premium, and 50 on GitLab.com Enterprise/Premium with Orbit, covering projects, issues, MRs, pipelines, CI/CD, and more
The N+1 fix changed the query to return all results without applying the default pagination limit.
Suggested fix: Add .Limit(defaultPageSize) to the query in user_repository.go:87.
graph LR
A[You] -->|natural language| B[AI Client]
B -->|MCP protocol| C[GitLab MCP Server]
C -->|REST v4 + GraphQL| D[GitLab Instance]
D -->|JSON response| C
C -->|structured result| B
B -->|formatted answer| A
The server acts as a bridge: your AI client sends tool calls over the MCP protocol, the server translates them into GitLab REST API v4 or GraphQL requests, and returns the results in both structured JSON (for the AI) and formatted Markdown (for you).
