CVE-2025-8943: Flowise OS command remote code execution
Flowise's Custom MCPs feature runs OS commands, such as `npx` to start a local MCP server, without a sandbox. With no authentication configured, which was the default before 3.0.1, a network attacker can run arbitrary commands on the host. The package data lists flowise 3.0.5 and earlier as vulnerable and gives no patched version.
Published Aug 14, 2025 · Updated Aug 18, 2025 · Analysis Nov 1, 2026
What an attacker can do
Custom MCPs lets a user define a command that Flowise launches to run an MCP server. The command runs as an OS process with the privileges of the Flowise service, with no sandbox. Flowise has only minimal authentication and authorization, with no role-based access control. Per the advisory, before 3.0.1 a default install runs with no authentication unless it is configured.
That combination lets an unauthenticated attacker with network access send a command to the node-load-method/customMCP API endpoint, which JFrog's proof of concept uses, and have it executed. The attacker gets full code execution on the Flowise host: its environment variables, the API keys and database credentials stored for flows, uploaded data, and a foothold on the internal network. Anyone who is authenticated but low-privileged can do the same, since there are no roles.
A public proof of concept exists. The EPSS score is very high (0.752). The sources do not say whether attacks in the wild were seen.
Who is affected
The npm package flowise at 3.0.5 and earlier, per the package data. JFrog lists the range without bounds, and the advisory text says 3.0.1 changed the default for authentication. Check:
npm ls -g flowiseIf you run the Docker image, check its tag. Instances reachable from untrusted networks, and instances with no FLOWISE_USERNAME/password-style authentication set, are the most exposed.
flowise
- Ecosystem
- npm
- Affected
- <= 3.0.5
- Fixed in
- No patched version yet
- Check
npm ls flowise
How to fix it
No patched version is listed. The package data shows patched empty, and the JFrog page says no mitigations are supplied. The advisory notes that Flowise 3.0.1 and later no longer default to running without authentication, which closes the unauthenticated path, though not the issue of authenticated users running commands.
Practical steps, from the advisory's root cause: turn on authentication, do not expose Flowise to the internet or untrusted networks, and run it as an unprivileged user in a container with limited access. Watch the Flowise releases for a version that removes or sandboxes Custom MCP command execution.
How to tell if you were hit
Look in web server and Flowise access logs for requests to node-load-method/customMCP, especially from unfamiliar IPs. On the host, check for child processes of the Flowise node process that you did not expect (shells, curl, wget, npx with odd packages), new files in temp directories, and outbound connections. Review custom MCP definitions saved in Flowise for commands you did not add.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
The Custom MCPs feature is designed to execute OS commands, for instance, using tools like npx to spin up local MCP Servers. However, Flowise's inherent authentication and authorization model is minimal and lacks role-based access controls (RBAC). Furthermore, in Flowise versions before 3.0.1 the default installation operates without authentication unless explicitly configured. This combination allows unauthenticated network attackers to execute unsandboxed OS commands.
Questions
- Is there a patched version of Flowise for CVE-2025-8943?
None is listed. The package data has no patched version and JFrog says no mitigations are supplied. Version 3.0.1 changed the default so installs no longer run without authentication.
- Does the attacker need to log in?
Not on installs running without authentication, which was the default before 3.0.1. Flowise has no role-based access, so any authenticated user could also reach the feature.
- Is there public exploit code?
Yes. The JFrog write-up includes a proof of concept that sends a command to the
node-load-method/customMCPendpoint. The sources do not state whether it is used in attacks in the wild.
References
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology
More Flowise vulnerabilities
All Flowise entriesCVE-2026-73483
Flowise NodeVM sandbox escape via puppeteer allowlist - authenticated RCE and arbitrary file read via Chromium
critical9.4Oct 7, 2026CVE-2026-73487
Flowise Prompt Injection to RCE and SSRF via CSV/Airtable Agent Python Validator Bypass
critical9.0Oct 7, 2026CVE-2026-69250
Flowise: Unauthenticated OAuth2 Refresh Enables Non-Blind SSRF and Secret Exfiltration
high8.5Aug 4, 2026CVE-2026-69251
Flowise RCE via TypeORM DataSource
critical9.0Aug 4, 2026CVE-2026-69252
Flowise: Missing authorization on `/api/v1/files` allows low-privileged API keys to list and delete files across workspaces within the same organization
high7.2Aug 4, 2026CVE-2026-69253
Flowise Sandbox Escape to RCE
critical9.0Aug 4, 2026
