Protect Your Tracking Server from Network Exposure

MLflow 3.5.0+ includes security middleware to protect against DNS rebinding, CORS attacks, and clickjacking. These features are available with the default FastAPI-based tracking server (uvicorn).

DNS Rebinding Protection

Validates Host headers to prevent attacks on internal services

CORS Protection

Controls which web apps can access your MLflow API

Clickjacking Prevention

X-Frame-Options header controls iframe embedding

Security Headers

Automatic headers prevent MIME sniffing and XSS

Security settings can be configured through CLI options or environment variables. Default settings are designed to be safe for local development while requiring explicit configuration for production:

Requirements

Security middleware features require the FastAPI-based tracking server (uvicorn), which is the default server in MLflow 3.5.0+.

These features are not available when using --gunicorn-opts or --waitress-opts.

Allowed Hosts

Controls which Host headers the server accepts. This prevents DNS rebinding attacks by validating incoming requests:

CLI Option	Environment Variable	Default
--allowed-hosts	MLFLOW_SERVER_ALLOWED_HOSTS	localhost, private IPs

Example:

# Specific hosts
mlflow server --allowed-hosts "mlflow.company.com,192.168.1.100"

# Wildcard patterns
mlflow server --allowed-hosts "*.company.com,192.168.*"

# Allow all (not recommended)
mlflow server --allowed-hosts "*"

CORS Origins

Specifies which web applications can make API requests from browsers:

CLI Option	Environment Variable	Default
--cors-allowed-origins	MLFLOW_SERVER_CORS_ALLOWED_ORIGINS	localhost:* (any localhost origin)

# Specific origins
mlflow server --cors-allowed-origins "https://app.company.com,https://notebook.company.com"

# Wildcard for subdomains
mlflow server --cors-allowed-origins "https://*.company.com"

# Allow all origins (development only)
mlflow server --cors-allowed-origins "*"

X-Frame-Options

Sets the X-Frame-Options header to control iframe embedding behavior:

CLI Option	Environment Variable	Default
--x-frame-options	MLFLOW_SERVER_X_FRAME_OPTIONS	SAMEORIGIN

• SAMEORIGIN - Only same origin can embed (default)
• DENY - No embedding allowed
• NONE - Any site can embed

# Allow cross-origin iframe embedding
mlflow server --x-frame-options NONE

Disable Security Middleware

Completely disables security middleware. Use this only when security is handled by a reverse proxy or gateway:

CLI Option	Environment Variable	Default
--disable-security-middleware	MLFLOW_SERVER_DISABLE_SECURITY_MIDDLEWARE	false

mlflow server --disable-security-middleware

Common Scenarios

Local Development

Default configuration works out of the box:

mlflow server

This setup accepts connections from localhost and any private IPs.

Remote Tracking Server

For a shared server accessed by multiple known users, you can configure it to allow connections from specific hosts.

mlflow server --host 0.0.0.0 --allowed-hosts "mlflow.internal:5000,localhost:*"

Then use MLflow Python SDK to connect to remote tracking servers:

import mlflow

# Connect to remote server
mlflow.set_tracking_uri("http://mlflow.company.com:5000")

with mlflow.start_run():
    mlflow.log_param("alpha", 0.5)
    mlflow.log_metric("rmse", 0.1)

Host vs Allowed Hosts

Render MLflow UI within Jupyter Notebooks

When notebooks run on a different domain than your MLflow server, configure CORS and frame options:

# Allow embedding from notebook domain
mlflow server --host 0.0.0.0 \
  --x-frame-options NONE \
  --cors-allowed-origins "https://jupyter.company.com"

Then embed specific MLflow views directly in notebook cells:

from IPython.display import IFrame

# Embed MLflow UI
IFrame(src="http://mlflow.company.com:5000", width=1000, height=600)

Embedding MLflow UI in Web Applications

Create a component to display MLflow content:

function MLflowDashboard() {
  return (
    <iframe
      src="http://mlflow.company.com:5000/experiments/1"
      style={{ width: '100%', height: '800px' }}
      title="MLflow"
    />
  );
}

Configure the MLflow server to accept requests from your webapp:

mlflow server --host 0.0.0.0 \
  --x-frame-options NONE \
  --cors-allowed-origins "http://localhost:3000,https://app.company.com"

Use this HTML file to verify iframe configuration works correctly:

<!DOCTYPE html>
<html>
<head>
    <title>MLflow iframe Test</title>
</head>
<body>
    <h1>MLflow Embedding Test</h1>
    <iframe
        src="http://localhost:5000"
        style="width: 100%; height: 600px; border: 1px solid #ccc;"
        onload="document.getElementById('status').innerHTML = '✅ Loaded'"
        onerror="document.getElementById('status').innerHTML = '❌ Failed'">
    </iframe>
    <div id="status">Loading...</div>
</body>
</html>