CVE-2026-105697: Langflow: OS command injection (RCE) via arbitrary command in MCP stdio server configuration

In MCPStdioClient._connect_to_server (lfx/base/mcp/util.py), the command string was wrapped in bash -c "exec {command_str} || echo ...". The MCP SDK starts that process before any MCP handshake, so the command runs even when the UI later reports that the stdio server failed to start.

The entry points are "Settings → MCP Servers → Add MCP Server", POST or PATCH /api/v2/mcp/servers/{server_name}, and the MCP Tools component. The reporter's proof of concept sent {"command":"touch","args":["/tmp/pwned_langflow"]} with a token from GET /api/v1/auto_login. The file appeared on the server. A real attacker would run a reverse shell, which runs as the Langflow process user and exposes stored API keys and anything the host can reach.

The fix came in two steps. Version 1.9.0 added an allowlist to the REST model only. Configs embedded in flows or passed as tweaks still reached the unguarded sink on 1.9.0 through 1.10.2, so those releases are not safe.

Affected: langflow >= 1.1.2 and < 1.10.3, langflow-base >= 0.1.2 and < 0.10.3, and lfx below 1.10.3. The MCP Stdio component appeared in 1.1.2, and the settings page and API in 1.5.0. Exposure is highest on instances reachable over a network with LANGFLOW_AUTO_LOGIN=true, which the project documents as development-only.

pip show langflow langflow-base lfx

Upgrade to langflow 1.10.3, langflow-base 0.10.3 and lfx 1.10.3, or 1.11.0 and later:

pip install --upgrade langflow

From 1.10.3 one policy, validate_mcp_stdio_config, applies at the API, in flows and right before spawn, and the process is started without a shell. The allowed commands are bash, cmd, docker, node, npx, python, python3 and uvx.

If you cannot upgrade, set LANGFLOW_AUTO_LOGIN=false, keep Langflow off untrusted networks, and give accounts only to trusted users. On multi-tenant setups the advisory also recommends LANGFLOW_MCP_SERVER_ALLOWED_PACKAGES and the interpreter and Docker hardening flags, because npx and uvx can still run any package by default.

Review the stored MCP server configurations and API logs for POST or PATCH /api/v2/mcp/servers/ calls whose command is not a normal MCP launcher, such as touch, rm, curl, sh or bash -c. Look for auto_login requests from unfamiliar addresses and for child processes of the Langflow process, such as a shell or a reverse shell connection. On patched versions a blocked attempt logs MCPStdioSecurityError or Command 'touch' is not allowed for security reasons.

Summary

Before Langflow 1.10.3, the MCP stdio transport launched whatever command / args a user put in an MCP server configuration, with no allowlist and (before 1.10.3) wrapped in bash -c "exec {command} ...". Any user able to reach the MCP server settings ("Settings → MCP Servers → Add MCP Server", POST/PATCH /api/v2/mcp/servers/{server_name}) or to build a flow with the MCP Tools component could add a "server" whose command is an arbitrary OS command (touch, rm -rf, a reverse shell, ...). The command runs on the Langflow host as the Langflow process user as soon as Langflow tries to connect to the server (listing servers, loading tools, running the flow) — even when the UI then reports that the stdio server failed to start.

With the default LANGFLOW_AUTO_LOGIN=true, GET /api/v1/auto_login hands out a token without credentials, so on an exposed instance running the default configuration this is reachable without an account. AUTO_LOGIN is documented as a development-only setting; with it disabled, any authenticated (non-admin) user can exploit it.

The issue was fixed in two steps and is fully fixed in Langflow 1.10.3 (and 1.11.0+):

• 1.9.0 — #12290 added a command allowlist and argument/env validation to the REST model (MCPServerConfig), blocking the "Add MCP Server" vector reported here.
• 1.10.3 — #14036 applied the same policy at the execution sink (lfx.base.mcp.util), including configs embedded in flows / tweaks and the final pre-spawn boundary, and removed the bash -c wrapper (the process is now exec'd directly, without a shell).

Affected versions

| Package (PyPI) | Vulnerable | Patched |
| `langflow` | `>= 1.1.2, < 1.10.3` | `1.10.3` |
| `langflow-base` | `>= 0.1.2, < 0.10.3` | `0.10.3` |
| `lfx` | `< 1.10.3` | `1.10.3` |

| Release | State |
| `1.1.2` – `1.4.x` | MCP Stdio component (added in #5148) runs `StdioServerParameters(command=..., args=...)` from the component's `command` field with no validation. |
| `1.5.0` – `1.8.x` | "Add MCP Server" settings page and `/api/v2/mcp/servers` API added (#8388). Stored stdio configs are launched via `bash -c "exec {command_str} ..."` with no validation. The PoC below works as-is. |
| `1.9.0` – `1.10.2` | #12290: `POST`/`PATCH /api/v2/mcp/servers` reject commands outside the allowlist. The execution sink still has no validation and still uses `bash -c`, so configs that do not go through `MCPServerConfig` (for example an MCP Tools component value embedded in a flow or passed as a tweak) can still execute arbitrary commands. |
| `1.10.3`+ | #14036: one shared policy (`lfx.base.mcp.security.validate_mcp_stdio_config`) is enforced at the API, at flow execution and right before the process is spawned; no shell is used. **Fixed.** |

Details

Vulnerable sink (src/lfx/src/lfx/base/mcp/util.py, MCPStdioClient._connect_to_server, before 1.10.3):

server_params = StdioServerParameters(
    command="bash",
    args=["-c", f"exec {command_str} || echo 'Command failed with exit code $?' >&2"],
    env=env_data,
)

command_str is built from the user's command + args. The only "validation" before 1.9.0 was _validate_node_installation, which only checks that Node.js is installed when the command contains npx. The MCP SDK then starts the process with anyio.open_process, so the command runs before any MCP handshake — which is why it executes even if the UI shows "failed to start".

PoC (as reported, Langflow 1.5.0 – 1.8.x)

In "Add MCP Server" → STDIO, add:

Name      - Test
Command   - touch
Arguments - /tmp/pwn

Or through the API, using a token from auto_login (default configuration):

TOKEN=$(curl -s http://127.0.0.1:7860/api/v1/auto_login | python3 -c 'import sys,json;print(json.load(sys.stdin)["access_token"])')

curl -s -X POST 'http://127.0.0.1:7860/api/v2/mcp/servers/testing' \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  --data-raw '{"command":"touch","args":["/tmp/pwned_langflow"]}'

/tmp/pwned_langflow gets created on the server. On 1.10.3+ the request is rejected with Command 'touch' is not allowed for security reasons. Allowed commands: bash, cmd, docker, node, npx, python, python3, uvx, and the same payload fails at the sink (MCPStdioClient.connect_to_server("touch /tmp/pwned_langflow") raises MCPStdioSecurityError, and no file is created). Wrappers such as bash -c "touch ...", sh -c ..., python3 -c ... and node -e ... are also rejected.

Fix

• #12290 (1.9.0) — MCPServerConfig validators: command allowlist (node, python, python3, npx, uvx, docker, and cmd/sh/bash only to wrap one of those), shell-metacharacter and dangerous-keyword checks on arguments, env var blocklist (LD_PRELOAD, NODE_OPTIONS, PYTHONPATH, BASH_ENV, ...), and Docker isolation checks.
• #14036 (1.10.3) — moves the policy into lfx.base.mcp.security.validate_mcp_stdio_config and enforces it at every entry point (API, embedded flow / tweak configs, the deprecated MCP Stdio component and just before spawn); it also launches the process without a shell:

command_parts = shlex.split(command_str)
command, args = command_parts[0], command_parts[1:]
validate_mcp_stdio_config(command, args, env)   # final pre-spawn enforcement
server_params = StdioServerParameters(command=command, args=final_args, env=env_data)

Later hardening (1.11.0, #13530) adds optional stricter modes for multi-tenant deployments.

Workarounds (if you cannot upgrade)

• Set LANGFLOW_AUTO_LOGIN=false and do not expose Langflow directly to untrusted networks.
• Only give Langflow accounts to trusted users.

Hardening recommendations for 1.10.3+

The allowlist still lets npx / uvx run any package by default, which is how MCP servers are normally distributed. On multi-tenant deployments, operators should also set:

• LANGFLOW_MCP_SERVER_ALLOWED_PACKAGES — the exact packages npx / uvx may run.
• LANGFLOW_MCP_SERVER_INTERPRETER_HARDENING=true and LANGFLOW_MCP_SERVER_DOCKER_HARDENING=true.
• On 1.11.1+ (#14150), restricting custom code (LANGFLOW_ALLOW_CUSTOM_COMPONENTS=false, LANGFLOW_CUSTOM_COMPONENT_ADMIN_ONLY=true or LANGFLOW_BLOCK_CODE_INTERPRETER_COMPONENTS=true) also limits MCP stdio servers to superusers.

Impact

Remote code execution on the Langflow host as the Langflow process user (CWE-78). Any exposed instance on a vulnerable version is affected, whether the attacker has an account or uses the default AUTO_LOGIN. The reporter found several internet-facing Langflow servers that were vulnerable.