Get Started

Real-time guardrail protection for AI agents and MCP endpoints

Deploy real-time protection for your AI agents and MCP servers. Akto Guardrails blocks threats as they occur, preventing prompt injections, tool abuse, and unauthorized access before they impact your systems.

Quick Start

Step 1: Sign Up or Log In

1. Go to app.akto.io

2. Sign up for a new account or log in to your existing account

Step 2: Navigate to Connectors

1. Go to Quick Start in your Akto dashboard

2. Scroll to the Guardrails section

3. Choose the proxy type that matches your setup

Step 3: Configure Your Proxy

Select and configure the appropriate protection proxy:

For MCP Servers

Deploy real-time protection for Model Context Protocol servers, tools, and resources.

→ Set Up MCP Proxy

What it protects:

• MCP tool invocations

• Resource access patterns

• Prompt handling

• Schema compliance

For AI Agents

Deploy real-time protection for AI agents and LLM applications.

→ Set Up Agent Proxy

What it protects:

• Agent requests and responses

• Tool permission enforcement

• Authentication patterns

• Data leakage prevention

Step 4: Enable Protection

Once your proxy is configured:

1. Set environment variable: AKTO_THREAT_ENABLED=true

2. Start routing traffic through the proxy

3. Monitor threats in Guardrails dashboard

4. Configure custom policies as needed

Step 5: Monitor & Respond

From your Guardrails dashboard:

1. View real-time guardrail detection

2. See blocked attacks and patterns

3. Configure blocking rules

4. Set up alerts for critical violations

What Gets Protected

Akto Guardrails defends against:

• Prompt injection attacks

• Tool abuse and permission escalation

• Unauthorized access attempts

• Schema violations

• Suspicious data patterns

• Known attack signatures

Key Features

• Real-Time Guardrails: Identify and respond to agentic security threats as they occur

• OWASP Core Rule Set Support: Detect common attack patterns including XSS and SQL injection in agent interactions

• Prompt Injection Protection: Block malicious prompts attempting to manipulate agent behavior

• Tool Permission Enforcement: Prevent unauthorized tool access and permission escalation

• Schema Conformance: Ensure agent requests comply with expected MCP schemas

• Geofencing: Restrict agent component access based on geographic locations

• IP-Based Blocking: Block or allow traffic from specific IP addresses or CIDR ranges

Next Steps

• Discover your agents to get full visibility

• Run Red Teaming Probes to identify vulnerabilities

• Configure custom blocking policies for your use case

Need help? Contact [e-mail] or use in-app Intercom.