Authenticated RCE in Langflow MCP Stdio Transport
CVE-2026-105740 is an OS command-injection vulnerability in Langflow versions earlier than 1.9.0. Any authenticated user, including a non-administrator, can add an MCP server using the Stdio transport and supply a command that Langflow passes directly to bash without validation, executable allowlisting, or sandboxing. Fetching the MCP server list triggers execution of the stored command with the privileges of the Langflow server process. The configuration's env field also permits arbitrary environment-variable injection, including dynamic-loader manipulation and executable search-path overrides. Stored malicious configurations can execute again whenever the server list is loaded. The vulnerability has a reported CVSS v3.1 score of 9.9 and is fixed in Langflow 1.9.0.
What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
Successful exploitation compromises confidentiality, integrity, and availability within the permissions and resources available to the Langflow process. Attackers can execute arbitrary commands, steal application secrets and credentials, read or modify accessible files, disrupt services, and establish persistent access through stored MCP configurations. Compromised credentials and network access can facilitate lateral movement. Exploitation attempts have included environment-variable theft and attempts to access the host filesystem through Docker; host-level compromise depends on deployment privileges and exposed resources.
If you can’t patch tonight, do this now.
Until upgrading, remove or disable MCP servers using the Stdio transport and prevent untrusted users from creating such configurations. Disable automatic login by setting LANGFLOW_AUTO_LOGIN=false, require strong unique credentials, restrict network access to trusted clients, and limit accounts to trusted users. Restricting administrator access alone is insufficient because ordinary authenticated users can exploit the vulnerability. Review MCP project settings explicitly, since malicious configurations may not appear during routine workflow-editor reviews. Command validation, executable allowlisting, and sandboxing provide additional defensive controls.
Patch, then assume compromise.
Upgrade Langflow to version 1.9.0 or later, which contains the fix for this vulnerability. Review stored MCP server configurations and authentication activity for evidence of exploitation, remove unauthorized configurations, and rotate exposed secrets and credentials if compromise is suspected.
No public exploits tracked yet. Mallory keeps watching.
27 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.