Threat Intel

CVE-2026-105697: Langflow MCP configuration enables shell execution

Unsafe MCP stdio configuration can execute commands in Langflow and lfx. The complete fix requires the patched package versions.

Armor11 min read

Tracking: CVE-2026-105697
Severity: Critical

Who is affected

The affected packages and first fixed releases are:

Package Affected versions First fixed release
langflow 1.1.2 to before 1.10.3 1.10.3
langflow-base 0.1.2 to before 0.10.3 0.10.3
lfx Before 1.10.3 1.10.3

With the default LANGFLOW_AUTO_LOGIN=true, a caller can obtain a token without supplying credentials. Disabling auto-login still leaves the reported path reachable to authenticated non-admin users.

How the attack works

MCP settings, flow execution, or execution tweaks can supply a stdio server command and arguments. A bash -c wrapper interprets that input as shell code. Execution happens before the MCP handshake, so a failed connection is not evidence that nothing ran.

The 1.9.0 API allowlist was a partial mitigation. The complete fix validates the execution path and starts the process directly.

What to do

  1. Upgrade every applicable package to the fixed release or later, then restart affected services.
  2. Disable unnecessary auto-login and review saved MCP commands, flow settings, and execution tweaks.
  3. Investigate unexpected child processes and failed MCP starts. Assess service credentials if suspicious execution occurred.

Sources

threat-intelmcpcvecommand-injection

See what your own fleet is running

Armor1 discovers and risk-scores every MCP server, skill and tool your developers have connected. Free for personal use, no credit card.

Start Free