AAK-MCP-LANGGRAPH-API-CVE-2026-118668-001: langgraph-api 0.1.2 to 0.14.x: a JS or TS auth handler's namespace rewrite is dropped
- Severity
- HIGH
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
LangGraphApiJsAuthStoreNamespace
What it detects
langgraph-api, the LangGraph agent server, runs JavaScript or TypeScript custom authentication in a Node.js sidecar. From 0.1.2 until 0.15.0 the server applied only a handler's changes to metadata back to the request, so a Store authorization handler that rewrote the requested namespace to keep each user in their own was ignored, and an authenticated user could read, overwrite, search or delete another user's Store items, the agents' long-term memory (CVE-2026-118668, CVSS 4.0 8.6, CWE-639; GHSA-4hm6-w6qq-w73v). Python handlers, and JS handlers that reject a foreign namespace rather than rewrite it, are not affected. Checked in the PyPI wheels: 0.1.0 has no JS auth bridge, 0.1.2 applies metadata only, and 0.15.0 applies every change the handler makes and refuses one it cannot apply. The newest 0.12.x, 0.13.x and 0.14.x releases, 0.14.10 of 2026-11-06 among them, do not carry the fix.
How to fix
Upgrade langgraph-api to 0.15.0 or later and pin it; a later 0.12, 0.13 or 0.14 release does not fix this. Until then, make every JS or TS Store authorization handler reject a request for a namespace the user does not own instead of rewriting it.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-LANGGRAPH-API-CVE-2026-118668-001
Limitations
Reports langgraph-api from 0.1.2 below 0.15.0 where a version is stated: a requirements*.txt entry at any depth, or a pyproject.toml, Pipfile or lockfile entry at the project root, reporting a range that admits an affected version. latest was 0.17.1 on 2026-11-11 and states no version, so an unversioned reference is not reported. It reads the version, not langgraph.json, so a deployment with Python custom auth, with none, or with JS handlers that reject rather than rewrite is reported too. The langchain/langgraph-api images are not read. A file over 2 MB, or a lockfile over 32 MiB, is not read. A declaration nested more than 64 levels deep in package.json, a TOML manifest or an MCP config (an MCP server more than 62) is still judged, and reported without a line; a lockfile entry keeps its line at any depth.
References
- CVE: CVE-2026-118668
- OWASP Agentic Top 10: ASI03 (Identity and Privilege Abuse), ASI06 (Memory & Context Poisoning)
- CSA AI Controls Matrix (AICM): IAM-01, STA-08
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-MCP-LANGGRAPH-API-CVE-2026-118668-001.md: the page then shows that text, with the registry reference appended under it.