Skip to content
NewsProduct launch

Claude Code 2.1.233 adds GitLab MR support, memory cgroup limits, and WebFetch cache control

· by Pondero Newsdesk

The short version

Anthropic's August 14 Claude Code release extends worktree and agents view to GitLab merge requests, adds Linux memory cgroup limits for Bash tool commands, and lets teams tune WebFetch cache duration.

Claude Code 2.1.233 adds GitLab MR support, memory cgroup limits, and WebFetch cache control

GitLab engineering teams got a long-requested feature on August 14: Claude Code 2.1.233 now understands GitLab merge request URLs in the worktree flag and the claude agents view, displaying them in the standard !N notation teams already use across their toolchain. The release bundled three other targeted additions for teams running Claude Code at scale.

What shipped

Per the Claude Code changelog, version 2.1.233 extended the worktree flag and agents view to handle GitLab merge request URLs. Merge requests now render as !N rather than as raw URLs, matching the convention GitLab teams rely on in issue trackers, CI pipelines, and code review tooling. Previous Claude Code releases treated GitLab MR URLs as opaque strings, which caused friction for teams running Claude Code against a GitLab remote.

The release also introduced memory cgroup support for Bash tool commands on Linux, enabled via the CLAUDE_CODE_TOOL_MEMORY_LIMIT environment variable. The feature is opt-in. Per the changelog, its purpose is to prevent a runaway build from stalling a session, a problem that surfaces when Claude Code manages parallel agent workloads or compilation-heavy tasks inside the Bash tool.

A new CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS environment variable controls how long fetched URLs stay in the session cache. The default remains 15 minutes. The variable exists so teams can shorten the window for frequently updated sources or extend it for stable reference pages.

The fourth addition targets enterprise gateway operators. An opt-in forward_user_identity setting on Anthropic upstreams sends the signed-in user's identity as request headers, so a proxy behind the gateway can attribute API spend to individual users rather than to the team account as a whole.

Why it matters

The GitLab MR change closes a concrete gap for the large share of engineering teams that host on GitLab rather than GitHub. The previous workaround required passing raw MR URLs or translating them manually; the agents view showed no !N prefix, which broke the familiar reference format that GitLab CI, code review, and project boards all rely on.

The memory cgroup feature is most relevant for Linux-based CI environments where Claude Code agents spawn build steps inside the Bash tool. Without a memory ceiling, a single unconstrained build job can exhaust available RAM on shared runners and stall neighboring sessions. Teams need to set CLAUDE_CODE_TOOL_MEMORY_LIMIT explicitly to activate the behavior; those that do not are unaffected.

forward_user_identity is the most actionable of the four for procurement-conscious engineering leaders. Per-user spend attribution lets organizations track which teams or individuals drive API consumption without routing requests through a separate accounting layer. It will matter most to larger organizations that already run a proxy or API gateway in front of Claude.

What to watch next

Whether GitLab CI pipeline triggers gain first-class support in a subsequent release is the natural next question, since that would extend the MR URL feature from interactive sessions into automated workflows. Teams operating multiple concurrent Claude Code sessions on shared Linux runners will also be early to validate whether the memory cgroup isolation performs as described in real build environments.

Sources