CVE-2026-44895
GitLab MCP Server lets an AI agent talk directly to GitLab
Published: 2026-05-26 · Last updated: 2026-06-01
Severity and scoring
- CWE
- CWE-306, CWE-942
Description
GitLab MCP Server lets an AI agent talk directly to GitLab. Prior to 0.6.0, the HTTP transport in src/transport.ts ships with no authentication layer at all and a wildcard Access-Control-Allow-Origin: * on every response. The structural defect is that the SSE server stands up a stateful, mutation-capable RPC endpoint that is backed by the operator's GITLAB_PERSONAL_ACCESS_TOKEN without any inbound credential check, then advertises itself to every cross-origin browser context via the wildcard CORS header. The httpServer.listen(port) call at line 97 also passes no host argument, so the bind defaults to 0.0.0.0 and exposes the auth-less surface on every interface. This vulnerability is fixed in 0.6.0.
Source: NVD
References
Related CVEs
Same CWE
- CVE-2026-50245 — Brickcom cameras allow unauthenticated access to live snapshot images via the /ONVIF endpoint and no authentication is required to retrie... (7.7 HIGH)
- CVE-2026-49973 — Hermes WebUI before version 0.51.358 contains an improper access control vulnerability that allows unauthenticated remote attackers to hi... (9.4 CRITICAL)
- CVE-2026-35273 — Vulnerability in the PeopleSoft Enterprise PeopleTools product of Oracle PeopleSoft (component: Updates Environment Management) (9.8 CRITICAL)
- CVE-2026-46612 — Fission is an open-source, Kubernetes-native serverless framework that simplifies the deployment of functions and applications on Kubernetes (8.8 HIGH)
- CVE-2026-20253 — In Splunk Enterprise versions below 10.2.4 and 10.0.7, and Splunk Cloud Platform versions below 10.4.2604.3 and 10.2.2510.14, an unauthen... (9.8 CRITICAL)