GitLab 19.5 is pushing agentic software development deeper into the core DevOps workflow, turning external AI agents from loosely connected tools into governed participants in the platform. Released on October 15, the update makes GitLab’s MCP server generally available and adds new controls for monitoring, restricting, and auditing how those agents are used.
GitLab 19.5 brings MCP access to the platform
The headline change in GitLab 19.5 is the general availability of the GitLab MCP server, which exposes GitLab capabilities through the Model Context Protocol. That lets MCP-compatible tools such as Claude Code and Cursor interact with GitLab’s broader tool catalogue and invoke GitLab Duo Agent Platform agents and flows.
According to the release, agents can now discover groups, inspect individual CI/CD artifact files, and start or monitor GitLab Duo agents and flows. Administrators can also decide which MCP toolsets are exposed, which GitLab says helps reduce both unnecessary access and the context agents consume.
That toolset selection matters because giving an agent every available capability increases the amount of context it must process and expands the number of actions it could take. GitLab frames the feature as a basic form of least-privilege design for agentic systems.
Governance expands to externally connected agents
One of the more notable additions is support for bringing Externally Connected Agents into GitLab’s AI Governance capabilities. Until now, the AI Governance Dashboard focused on GitLab Duo Agent Platform agents. With version 19.5, tools such as Claude Code, Cursor, and Codex can be connected through the GitLab CLI.
Once configured, organisations can record qualifying agent sessions and see the user, agent, project, and timing. That activity appears alongside GitLab’s own agents, and audit events can be streamed to a SIEM.
The release makes a clear distinction between connectivity and control. As the source material notes, connectivity creates capability; governance determines whether that capability can be safely used at organisational scale.
Rate limits and exposure controls
GitLab.com now applies MCP rate limits of 60 requests per minute for Free users and 600 for Premium and Ultimate users. Reconnects and retries count toward those limits.
The company is also giving administrators more control over what the connected agents can see and do. That includes the ability to restrict exposed toolsets, rather than opening the full MCP surface area to every agent by default.
GitLab Duo CLI gets a more autonomous mode
GitLab Duo CLI in 19.5 adds an auto mode that allows agents to execute tools without asking for approval for every shell command, Git operation, or MCP call. The feature is disabled by default, and administrators can control it at the instance, group, and project levels.
Some tools can still be blocked entirely if they are marked Always Deny. Combined with the new governance features, the change suggests GitLab is building for workflows where developers define the goal and agents carry out the steps under policy guardrails.
The new /goal command
The /goal command is now generally available as well. It lets developers describe an open-ended objective and have GitLab Duo delegate implementation and verification to a governed flow.
Taken together, these changes move GitLab closer to an operating model in which people supply intent while agents handle multi-step execution across the development environment.
Security and secrets controls also get stronger
GitLab 19.5 also makes GitLab Secrets Manager generally available. Built on OpenBao, it integrates with existing GitLab group and project permissions, scopes secrets to the jobs that need them, and generates audit events for secret operations.
The release further expands security configuration profiles. Organisations can centrally customise settings such as excluded paths, analyzer images, and full-history scans for Secret Detection and SAST, rather than repeating those settings across every project’s CI/CD file.
These updates are important because agentic development depends on access to repositories, credentials, and pipelines. GitLab is pairing broader agent capabilities with stronger controls around the systems those agents may touch.
Why this release matters
GitLab’s approach increasingly resembles the broader industry shift toward coding agents that do more than autocomplete lines of code. But the company is placing special emphasis on connecting external agents to an existing DevSecOps governance model, rather than treating them as separate developer tools.
- MCP makes GitLab’s platform accessible to external AI agents.
- Governance features provide visibility into who used an agent, on which project, and when.
- Auto mode and /goal point toward more autonomous multi-step development workflows.
- Secrets and security controls aim to keep that autonomy within managed boundaries.
In practical terms, the competitive question is no longer just which coding assistant writes better code. GitLab is betting that the bigger issue for enterprises is which platform can let autonomous agents work safely inside production development and compliance processes.
Source: Original report
Was this helpful?
Explore more: Software Development More Software Development Tech News
Last Modified: November 6, 2026 at 10:34 pm
4 views
