AAK-MCP-N8N-CVE-2026-72768-001: n8n below 2.41.4 / 2.42.0 (MCP workflow-validation prototype pollution, MCP Client SSRF bypass, node-schema loader RCE, cross-project credential reference, inline sub-workflow credential exfiltration)
- Severity
- HIGH
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
N8nMcpClientNodeSsrfBypass
What it detects
Five MCP-surface defects in n8n, carried by one pin because they share a package and the highest fix version covers all of them. Before 2.32.1 the MCP Client node bypasses server-side-request-forgery protection: an authenticated user crafts a workflow whose MCP Client requests reach internal or blocked hosts and reads their responses back through the workflow (CVE-2026-72768). Before 2.33.4, and on the 2.34.x line before 2.34.1, the @n8n/workflow-sdk node-schema loader used for MCP node-schema loading derives a schema module path directly from the attacker-supplied node type string with no path-traversal validation, so a global:member user reaches code execution in the n8n main process (CVE-2026-77068). Before 2.34.1 the MCP create_workflow_from_code tool skips credential validation when the authentication type is an expression, letting a holder of a valid MCP Bearer API key persist unauthorized cross-project credential references (CVE-2026-77073). Before 2.35.4, and on the 2.36.x line before 2.36.2, nodes that execute an inline sub-workflow (the Workflow Tool node) accept a credential reference in their inline workflow JSON without checking that the author owns it; MCP is one of the named write paths, so a shared-workflow editor persists a node naming somebody else's credential and the secret is resolved and exfiltrated when the workflow later runs under an identity that holds it (CVE-2026-85166). Before 2.41.4, and at 2.42.0, the MCP workflow-validation tool evaluates caller-supplied code in the main process through an interpreter that reads inherited properties and allows assignment onto shared prototypes, so a member holding only a workflow:read grant replaces a built-in method with one that returns a truthy value, permission checks then accept scopes it does not hold, and an owner account without MFA can be taken over (CVE-2026-110842, CVSS 4.0 7.7). That fix is the highest, so it sets the floor: 2.41.4, and 2.42.1 on the 2.42.x line, whose 2.42.0 sorts above 2.41.4 and is still affected. A project pinning n8n below 2.41.4, or on 2.42.0, or leaving it unpinned, is exposed to at least one of these. Distinct from the earlier n8n credential-domain-bypass and OAuth two-branch-fix pins the scanner already carries; the credential-domain pin's floor rose to the same 2.41.4 for a MEDIUM webhook bypass.
How to fix
Upgrade n8n to >= 2.41.4 and pin it, or to >= 2.42.1 if you are on the 2.42.x line: 2.42.0 is newer than 2.41.4 and still affected, so "at least 2.41.4" is not sufficient on that branch. Until then, give MCP access only to trusted members and turn on MFA for owner accounts. Route the MCP Client node's outbound requests through n8n's SSRF protection so a workflow author cannot reach internal or blocked hosts, and segment the n8n runner from sensitive internal services. Audit workflows for credential references that cross project boundaries, including inside the inline workflow JSON of sub-workflow nodes, which is where CVE-2026-85166 hides one.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-N8N-CVE-2026-72768-001
Limitations
Reads n8n in requirements*.txt at any depth and, at the project root only, in pyproject.toml, Pipfile, package.json, .mcp.json, mcp.json, claude_desktop_config.json and the Pipfile.lock, poetry.lock, uv.lock, package-lock.json, pnpm-lock.yaml and yarn.lock lockfiles, reporting a range that admits an affected version and a reference that states no version. The dependency is not read when it is declared in a sub-directory manifest (packages/app/package.json) or in a lockfile format the reader does not parse (a Yarn 2+ yarn.lock). A compose file's n8nio/n8n image tag is 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-2026-72768, CVE-2026-77068, CVE-2026-77073, CVE-2026-85166, CVE-2026-110842
- OWASP MCP Top 10: MCP01:2025 (Token Mismanagement & Secret Exposure), MCP04:2025 (Software Supply Chain Attacks & Dependency Tampering), MCP05:2025 (Command Injection & Execution)
- OWASP Agentic Top 10: ASI03 (Identity and Privilege Abuse), ASI04 (Agentic Supply Chain Vulnerabilities), ASI05 (Unexpected Code Execution (RCE))
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SSRF-01 (Internal Network MCP)
- CSA AI Controls Matrix (AICM): IVS-04, 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-N8N-CVE-2026-72768-001.md: the page then shows that text, with the registry reference appended under it.