GitLab has released a comprehensive walkthrough on integrating local coding harnesses like Claude Code with the GitLab Duo Agent Platform. Published on November 20, 2026, the guide demonstrates how developers can maintain their preferred local environment while offloading long-running, repository-aware tasks to GitLab’s agentic infrastructure. The core mechanism relies on a single MCP connection, allowing agents to trigger workflows for issue resolution, code review, pipeline repair, and security triage without leaving the terminal.

The Setup Protocol

The process begins with installing the GitLab Duo setup skill, which configures the user’s top-level group with necessary settings, flows, and automations. This skill is harness-neutral, supporting Claude Code, Codex, Cursor, and OpenCode, provided they are MCP-compatible. Once installed, the skill generates a plan that outlines every change to the group’s configuration, including turning on Duo Agent Platform workflows and connecting the GitLab MCP server. The setup requires Owner privileges on the top-level group to apply changes globally, though Maintainers can configure individual projects.

Four Key Workflows

The walkthrough details four primary use cases driven from the local terminal. First, developers can assign an issue to the Developer flow, which runs as a CI job on a GitLab runner to write code and open a merge request. Second, automated code reviews are enabled via the Duo Code Review flow, which scans merge requests and leaves inline feedback that local agents can read and resolve. Third, the Fix CI/CD Pipeline automation triggers when a pipeline fails, attempting to apply inline suggestions or open new MRs to fix the issue. Fourth, GitLab Ultimate users can leverage security flows to triage false positives and automatically resolve SAST vulnerabilities.

Governance and Context

GitLab emphasizes that agentic changes are subject to the same governance policies as human-authored code, including merge request approvals and pipeline execution rules. The integration also highlights the role of GitLab Orbit, a context graph generally available as of this update, which connects code, merge requests, and vulnerabilities to reduce agent retries. Internal testing reportedly showed that agents connected to Orbit required up to 45x fewer retries and used 4.5x fewer tokens. This context ensures that agents have a holistic view of the software lifecycle, preventing blind iterations on isolated merge requests.

Key Takeaways

  • Local harnesses like Claude Code act as the command center, while GitLab handles heavy lifting via CI runners.
  • The GitLab Duo setup skill automates group-wide configuration, including MCP connections and flow enablement.
  • Security workflows for SAST and secret detection are exclusive to GitLab Ultimate licensing.
  • GitLab Orbit integration significantly reduces token usage and retry loops by providing full lifecycle context.

The Bottom Line

This integration effectively bridges the gap between local development speed and enterprise-grade agentic automation. By keeping the human in the terminal loop while delegating heavy lifting to GitLab’s infrastructure, it offers a pragmatic path to scaling AI-assisted engineering without sacrificing governance or context.