GitLab Zero-Leak MCP

Example Use Cases

• 1

  Perform write operations on GitLab (create/update files, push commits, create branches, merge requests, and issues) while ensuring no arbitrary source-code reads occur.

• 2

  Integrate with Claude Desktop or other LLM-based agents to manage repositories securely with audit logging and a zero-read default policy.

• 3

  Controlled, allowlist-limited read access to specific public or documentation projects while blocking read access to all other repositories.

Description

This MCP server wraps the GitLab API with a zero-read default policy, disabling sensitive handlers (e.g., get_file_contents) and enabling environment-driven configuration for allowed read operations. It supports common repository operations (create/update files, push commits, branches, merge requests, issues, forks) while enforcing project-level allowlists and comprehensive audit logging. The project is Dockerized for local/self-hosted deployment and is designed for safe integration with Claude Desktop and other LLM-based agents. Recommended practices and environment variables are documented to minimize token and exposure risk.