AAK-MCP-MARIMO-CVE-2026-75149-001: marimo launches a notebook-supplied MCP server command on open (below 0.23.15)
- Severity
- HIGH
- Category
- Supply Chain (
supply-chain) - Scanner
agent_audit_kit/scanners/mcp_cve_pins_2026_07.py- Auto-fixable
- No.
- SARIF name
MarimoNotebookMcpCommandInjection
What it detects
marimo before 0.23.15 has code injection in the notebook configuration handler: a notebook can embed an MCP server entry whose command value is attacker-controlled, and opening that notebook in edit mode launches the command as a local subprocess before any cell executes and without authentication (CVE-2026-75149, CVSS 3.1 8.8, CWE-94). Cloning a repository and opening the notebook is the whole exploit. This is the stdio-launcher shape AAK-MCP-STDIO-LAUNCHER-INJECT-001 already describes, arriving through a notebook file rather than an MCP config, which is why it needs a pin: the config scanners never open a .py notebook. Fixed in 0.23.15. The floor carries no lower bound because two further pip-ecosystem marimo advisories record < 0.23.0 (pre-auth RCE) and < 0.23.9 (reflected XSS), so anything below this floor is exposed to something.
How to fix
Upgrade marimo to >= 0.23.15 and pin it. Until then, do not open untrusted notebooks in edit mode: the subprocess launches before any cell runs, so reviewing the cells first does not protect you. Treat a notebook from an untrusted repository as executable content, not as a document.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-MCP-MARIMO-CVE-2026-75149-001
Limitations
Reads marimo in requirements*.txt at any depth and, at the project root only, in pyproject.toml, Pipfile, .mcp.json, mcp.json, claude_desktop_config.json and the Pipfile.lock, poetry.lock and uv.lock lockfiles, taking the lowest version a range allows and reporting a reference that states no version. The dependency is not read when it is declared in a sub-directory manifest (services/api/pyproject.toml) or in a lockfile format the reader does not parse (pdm.lock). 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-2026-75149
- OWASP MCP Top 10: MCP04:2025 (Software Supply Chain Attacks & Dependency Tampering), MCP05:2025 (Command Injection & Execution)
- OWASP Agentic Top 10: ASI04 (Agentic Supply Chain Vulnerabilities), ASI05 (Unexpected Code Execution (RCE))
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-INJECT-01 (Shell Injection via 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-MARIMO-CVE-2026-75149-001.md: the page then shows that text, with the registry reference appended under it.