Any logged-in Langflow user can run OS commands on your server by adding an MCP server with the stdio transport — the command field is passed straight to the shell with no validation. Security researchers have logged 1,423 exploit attempts against honeypots since the disclosure. This free 30-second self-test tells you if you're at risk. No signup, nothing installed.
Check your Langflow version — on the host running Langflow:
pip show langflow | grep -i version
Docker users: check the image tag you deployed (e.g. langflowai/langflow:1.8.x).
Audit your MCP servers — list every configured MCP server and look for stdio-transport entries you didn't authorize:
curl -s -H "Authorization: Bearer $LF_API_KEY" \
http://localhost:7860/api/v2/mcp/servers | python3 -m json.tool | grep -A3 stdio
Any stdio entry with a command you don't recognize was potentially planted — the payload executes when the server list is fetched.
Check for signs of past compromise — look for unexpected MCP server registrations in your logs:
grep "POST /api/v2/mcp/servers" /var/log/langflow/*.log | grep -v "your-admin-user"
Also hunt for new processes, cron entries, or outbound connections you can't explain — RCE gives the attacker the whole host, not just Langflow.
This page checked one CVE. Rogue-Agent Defense monitors your AI agents for prompt injection, data exfiltration, and anomalous tool use — the things an exploited agent does next.
Drop a live version-checker into your blog, docs, or forum post — three lines of HTML.
Get the embed code