GitLab MR MCP

Connect your AI assistant to GitLab. Ask questions like "List open merge requests", "Show me reviews for MR #123", "Get commit discussions for MR #456", or "Find merge requests for the feature branch" directly in your chat.

Table of Contents

• Quick Setup

• What You Can Do

• Configuration Options

• Troubleshooting

• Tool Reference

• Roadmap

• Development

• Security Notes

• Support

Related MCP server: GitHub MCP Tools

Quick Setup

Installation

Note: Using pipx or uv tool is recommended as they automatically add the gitlab-mcp command to your PATH. If using pip install, ensure your Python scripts directory is in PATH, or use the full path to the command.

Get your GitLab token

1. Go to GitLab → Settings → Access Tokens

2. Create token with read_api scope (add api scope if you want write access)

3. Copy the token

Configure your MCP client

Multi-Project Setup (Recommended)

For working with multiple GitLab projects, add to your global MCP config (~/.cursor/mcp.json for Cursor):

This single configuration works across all your projects. Use search_projects or list_my_projects to find project IDs, then specify project_id in your requests.

Single-Project Setup

For working with a single project, you can set a default project ID:

Restart your MCP client and start asking GitLab questions!

What You Can Do

Once connected, try these commands in your chat:

Multi-Project Workflow

• "What projects do I have access to?"

• "Search for the backend project"

• "Show open MRs for project 12345"

• "List merge requests for group/my-project"

Single-Project Commands

• "List open merge requests"

• "Show me details for merge request 456"

• "Get reviews and discussions for MR #123"

• "Show me the test summary for MR #456"

• "What tests failed in merge request #789?"

• "Show me the pipeline for MR #456"

• "Get the failed job logs for merge request #789"

• "Show me commit discussions for MR #456"

• "Get all comments on commits in merge request #789"

• "Find merge requests for the feature/auth-improvements branch"

• "Show me closed merge requests targeting main"

• "Reply to discussion abc123 in MR #456 with 'Thanks for the feedback!'"

• "Create a new review comment in MR #789 asking about the error handling"

• "Resolve discussion def456 in MR #123"

• "Approve merge request #456"

• "Merge MR #123 with squash"

• "Merge MR #789 when pipeline succeeds"

Working with Review Comments

The enhanced review tools allow you to interact with merge request discussions:

1. First, get the reviews to see discussion IDs:

   "Show me reviews for MR #123"

2. Reply to specific discussions using the discussion ID:

   "Reply to discussion abc123 in MR #456 with 'I'll fix this in the next commit'"

3. Create new discussion threads to start conversations:

   "Create a review comment in MR #789 asking 'Could you add error handling here?'"

4. Resolve discussions when issues are addressed:

   "Resolve discussion def456 in MR #123"

Note: The get_merge_request_reviews tool now displays discussion IDs and note IDs in the output, making it easy to reference specific discussions when replying or resolving.

Approving and Merging

Complete the MR lifecycle with approval and merge tools:

1. Approve a merge request:

   "Approve MR #123"

2. Merge with options:

   "Merge MR #456 with squash" "Merge MR #789 and remove source branch" "Merge MR #123 when pipeline succeeds"

3. Revoke approval (if needed):

   "Unapprove MR #456"

Merge Options:

• squash - Squash commits into a single commit

• should_remove_source_branch - Delete source branch after merge

• merge_when_pipeline_succeeds - Auto-merge when pipeline passes

• sha - Ensure HEAD hasn't changed (safety check)

Note: You cannot approve your own MRs. The merge will fail if the MR has conflicts, is in draft status, or doesn't meet approval requirements.

Working with Test Reports (Recommended for Test Failures)

GitLab provides two tools for checking test results - use the summary for quick checks, and the full report for detailed debugging:

Option 1: Test Summary (Fast & Lightweight) ⚡

Use get_pipeline_test_summary for a quick overview:

What You Get:

• 📊 Pass/fail counts per test suite

• ⏱️ Total execution time

• 🎯 Pass rate percentage

• ⚡ Fast - doesn't include detailed error messages

Option 2: Full Test Report (Detailed) 🔍

Use get_merge_request_test_report for detailed debugging:

What You Get:

• ✅ Specific test names that passed/failed

• ❌ Error messages and stack traces

• 📦 Test suites organized by class/file

• ⏱️ Execution time for each test

• 📊 Pass rate and summary statistics

• 📄 File paths and line numbers

How Both Work:

• Automatically fetch the latest pipeline for the merge request

• Retrieve test data from that pipeline (uses GitLab's /pipelines/:pipeline_id/test_report or /test_report_summary API)

Example Output:

Why Use This Instead of Job Logs?

• 🎯 No noise: Only test results, no build/setup output

• 📊 Structured data: Easy for AI to understand and suggest fixes

• 🚀 Fast: Much smaller than full job logs

• 🔍 Precise: Shows exact test names and error locations

Requirements:

Your CI must upload test results using artifacts:reports:junit in .gitlab-ci.yml:

Working with Pipeline Jobs and Logs

The pipeline tools provide a two-step workflow for debugging test failures:

Step 1: Get Pipeline Overview

Use get_merge_request_pipeline to see all jobs and their statuses:

What You Get:

• Pipeline overview (status, duration, coverage)

• All jobs grouped by status (failed, running, success)

• Job IDs for each job (use these to fetch logs)

• Direct links to view jobs in GitLab

• Job-level timing and stage information

Step 2: Get Specific Job Logs

Use get_job_log with a job ID to fetch the actual output:

What You Get:

• Complete job output/trace

• Log size and line count

• Automatically truncated to last 15,000 characters for very long logs

Typical Workflow:

Why Two Tools?

• Performance: Only fetch logs when needed (not all at once)

• Flexibility: Check any job's log (failed, successful, or running)

• Context Efficient: Avoid dumping huge logs unnecessarily

Working with Commit Discussions

The get_commit_discussions tool provides comprehensive insights into discussions and comments on individual commits within a merge request:

1. View all commit discussions for a merge request:

   "Show me commit discussions for MR #123"

2. Get detailed commit conversation history:

   "Get all comments on commits in merge request #456"

This tool is particularly useful for:

• Code Review Tracking: See all feedback on specific commits

• Discussion History: Understand the evolution of code discussions

• Commit-Level Context: View comments tied to specific code changes

• Review Progress: Monitor which commits have been discussed

Technical Implementation:

• Uses /projects/:project_id/merge_requests/:merge_request_iid/commits to get all commits with proper pagination

• Fetches ALL merge request discussions using /projects/:project_id/merge_requests/:merge_request_iid/discussions with pagination support

• Filters discussions by commit SHA using position data to show commit-specific conversations

• Handles both individual comments and discussion threads correctly

The output includes:

• Summary of total commits and discussion counts

• Individual commit details (SHA, title, author, date)

• All discussions and comments for each commit with file positions

• Complete conversation threads with replies

• File positions for diff-related comments

• Thread conversations with replies

Configuration Options

MCP Config (Recommended)

Configure environment variables directly in your MCP client config as shown in Quick Setup. This keeps project-specific settings with the project.

Environment Variables

Alternatively, set environment variables in your shell:

SOCKS Proxy Support

Route all GitLab API requests through a SOCKS5 proxy by setting SOCKS_PROXY:

Or via environment variable:

When SOCKS_PROXY is not set, connections are made directly (no proxy).

Find Your Project ID

• Go to your GitLab project → Settings → General → Project ID

• Or check the URL: https://gitlab.com/username/project (use the numeric ID)

Troubleshooting

Authentication Error: Verify your token has read_api permissions and is not expired.

Project Not Found: Double-check your project ID is correct (it's a number, not the project name).

Connection Issues: Make sure your GitLab URL is accessible and correct.

Script Not Found: Ensure the path in your MCP config points to the actual server location and the script is executable.

Tool Reference

Project Discovery Tools

Merge Request Tools

All project-scoped tools accept an optional project_id parameter. If not provided, falls back to GITLAB_PROJECT_ID env var.

Roadmap

Recently Added

• v1.4.0: Project discovery tools, MCP best practices (tool titles, annotations), improved prompts

• v1.3.1: Fixed multi-workspace environment variable conflict in Cursor

• v1.3.0: SOCKS5 proxy support for routing GitLab API requests

• v1.2.0: Merge, approve, and unapprove MR tools - complete MR lifecycle

• v1.1.0: Create and update MR tools, cleaner output formatting

Coming Next

• Issue Management - List, create, update issues and add comments

• Inline Comments - Add code review comments on specific lines

Considering

• Lightweight file list for MRs (changed files without full diff)

• Rebase MR via API

Out of Scope

Branch operations, file content fetching, and full diffs are intentionally not included - use git locally for these tasks, it's faster and more capable.

Have a feature request? Open an issue!

Development

Project Structure

Adding Tools

1. Create new file in gitlab_mr_mcp/tools/ directory

2. Add import and export to gitlab_mr_mcp/tools/__init__.py

3. Add to list_tools() in gitlab_mr_mcp/server.py

4. Add handler to call_tool() in gitlab_mr_mcp/server.py

Adding Prompts

Prompts provide workflow guidance to AI assistants. Add new prompts in gitlab_mr_mcp/prompts.py:

1. Define the prompt content as a string constant

2. Add entry to the PROMPTS dictionary with title, description, and content

Development Setup

1. Install development dependencies:

2. Available make commands:

3. Set up pre-commit hooks:

This will automatically check and format your code for:

• ✨ Trailing whitespace - auto-removed

• 📄 End-of-file issues - auto-fixed

• 🎨 Code formatting (black) - auto-formatted

• 📦 Import sorting (isort) - auto-organized

• 🐍 Python style (flake8) - linted with bugbear & print detection

• 🔒 Security issues (bandit) - security checks

• 📋 YAML/JSON formatting - validated

4. Format all existing code (first time only):

5. Run pre-commit manually on all files:

Running Tests

Security Notes

• Never commit access tokens to version control

• Use project-specific tokens with minimal permissions (read_api scope)

• Rotate tokens regularly

• Store tokens in your MCP config (which should not be committed)

Support

• Check GitLab API documentation

• Open issues at github.com/amirsina-mandegari/gitlab-mr-mcp

License

MIT License - see LICENSE file for details.