GitLab has made agentic software development a governed part of its DevOps platform with the release of GitLab 19.5 on 15 October. The update makes the GitLab MCP server generally available, so AI coding tools such as Claude Code and Cursor can reach GitLab's wider tool catalogue and invoke GitLab Duo Agent Platform agents and flows from inside the tools developers already use.

The generally available MCP server, built on the Model Context Protocol, gives agents access to GitLab capabilities with new tools to discover groups, inspect individual CI/CD artifact files, and start or monitor Duo agents and flows. Administrators can choose which MCP toolsets are exposed, cutting down unnecessary tools and the context they consume, and GitLab.com now applies MCP rate limits of 60 requests a minute for Free users and 600 for Premium and Ultimate users.

A significant addition is governance. Externally connected agents, including Claude Code, Cursor and Codex, can now be brought into GitLab's AI Governance capabilities through the GitLab CLI, giving organisations visibility into what outside agents do with their code, issues and pipelines. GitLab Duo CLI also gains an auto mode, letting agents run tools without asking approval for every shell command, Git operation or MCP call.

GitLab Secrets Manager is now generally available too. Built on OpenBao, it integrates with existing GitLab group and project permissions and scopes secrets to the jobs that need them, so agents and pipelines no longer need broad credentials to do their work.

The release lands as AI coding assistants spread fast across development teams, raising the question of who watches the watchers. By putting external agents under the same governance as its own Duo platform, GitLab is betting that control, not just capability, will decide which agentic tools enterprises trust. InfoQ covers the release in detail.