AVAILABLE NOW — Published on npm

Your AI Agents Are Running Blind.
We See Everything.

5-layer defense-in-depth. 60 threat patterns. Prompt injection detection.
Token cost monitoring. Self-healing security. Enterprise-ready.

$ npx autoai-agentshield
Install Now — Free → See it in action ↓ View Product Poster
60 threat patterns
•
5 defense layers
•
10 PII detectors
•
45 tests passing
PRODUCT DEMO

See AgentShield in Action

42-second commercial showing the problem, the product, and the proof.

HOW IT WORKS

How AgentShield Protects Your Agents

When connected via MCP, AgentShield intercepts every tool call through 5 defense layers. Below is an illustration of what happens internally when requests flow through AgentShield.

AgentShield Defense Layers (via MCP) ACTIVE
ARCHITECTURE

5-Layer Defense-in-Depth

Every agent request passes through five independent security layers. An attacker must defeat all five to succeed.

1
Input Scanning
Detect prompt injections before they reach the agent
▾
  • 13 injection pattern signatures
  • Shannon entropy analysis
  • Homoglyph detection
  • Base64/hex payload scanning
  • Recursive prompt unwrapping
  • Context window overflow detection
2
Identity & Authorization
Verify who the agent is and what it can do
▾
  • Role-based access control (RBAC)
  • AWS Cognito integration
  • Azure Entra ID (AAD)
  • Custom JWT validation
  • Agent identity attestation
  • Session-scoped permissions
3
Policy Engine
Deterministic rules that cannot be bypassed by AI
▾
  • Deterministic policy evaluation
  • Cedar policy language support
  • Resource-level permissions
  • Time-based access windows
  • Rate limiting per agent/tool
  • Cost budget enforcement
4
Output Validation
Catch sensitive data before it leaves your system
▾
  • 10 PII pattern detectors
  • Credit card / SSN scanning
  • API key & secret detection
  • Custom regex patterns
  • Content toxicity filtering
  • Data classification tagging
5
Observability & Audit
Complete audit trail for compliance and forensics
▾
  • Full request/response logging
  • Structured JSON audit events
  • SIEM integration (Splunk, Datadog)
  • Real-time alerting
  • Compliance report generation
  • Forensic replay capability
WHAT NOBODY ELSE HAS

Innovations That Change the Game

Five capabilities you won't find in any other agent security tool.

🌀

Overthinking Loop Detection

Catches 142x token amplification attacks where adversarial prompts cause agents to spiral, burning your budget.

Monitors token consumption in real-time. When an agent's output exceeds the expected response length by a configurable multiplier (default: 5x), AgentShield interrupts execution, logs the anomaly, and returns a safe fallback response. Prevents runaway costs from recursive reasoning attacks.
🛡

MCP Server Trust Registry

Security scores for every MCP server your agents connect to. Know what you're trusting before you trust it.

Maintains a continuously updated registry of MCP servers with security scores based on: code audit status, maintainer reputation, dependency chain analysis, known vulnerabilities, and behavioral monitoring. Blocks connections to servers below your trust threshold.
🔗

Cross-Agent Threat Correlation

Connects attack chains across multiple agents. See the full kill chain, not just isolated alerts.

When an attacker probes agent-A to discover information, then uses that to exploit agent-B, AgentShield correlates these events using session tracking and behavioral fingerprinting. Produces unified incident reports showing the complete attack narrative.
⚙

Self-Healing Security

Auto-isolate compromised agents. Auto-rotate credentials. Auto-revoke sessions. Zero human intervention.

When a breach is detected, AgentShield executes a configurable response playbook: isolate the agent from all tools, rotate affected API keys, revoke active sessions, alert the team via Slack/PagerDuty, and generate a forensic report. Median response time: under 30 seconds.
🔍

Supply Chain Verification

Like npm audit, but for MCP servers. Know every dependency your agents rely on and its security posture.

Scans the full dependency tree of every MCP server: npm packages, Docker base images, API endpoints called, and data flows. Generates a Software Bill of Materials (SBOM) for your agent ecosystem. Alerts on newly discovered CVEs in any component.
INTEGRATIONS

Plugs Into Your Existing Stack

Native integrations with the platforms your enterprise already uses.

Cloud & Identity
☁ AWS Bedrock 🔐 AWS Cognito 🌳 AWS Cedar ☁ Azure Entra ID 🛡 Azure Content Safety ☁ GCP IAM 🤖 GCP Vertex AI
SIEM & Observability
📊 Splunk 🐶 Datadog 🔎 Elastic 🛡 Microsoft Sentinel 📜 Chronicle SIEM
Escalation & Alerting
💬 Slack 👥 Microsoft Teams 🚨 PagerDuty ✉ Email (SMTP) 📱 Webhooks
REAL-WORLD IMPACT

Before vs. After AgentShield

The measurable difference between unprotected agents and shielded ones.

Metric Without AgentShield With AgentShield
Prompt injection detection0% (unmonitored)94% (60 patterns)
Time to detect data leakDays / never<1 second
Credential rotation after breachManual, hoursAutomatic, <30s
Compliance audit trailSpreadsheetsComplete, queryable
Agent trust verificationNoneContinuous scoring
Token cost anomaly detectionMonthly bill shockReal-time alerts
MCP server supply chain auditNot possibleAutomated SBOM
SECURITY & TRUST

Verify, Don't Trust

Six guarantees we make about how AgentShield handles your data.

🏠 Runs Locally

AgentShield runs entirely on your machine. No cloud calls, no proxy servers, no data leaves your network.

Verify: check network tab — zero outbound requests

🔓 Open Source

Every line of code is on GitHub. Audit it yourself, fork it, contribute to it.

Verify: github.com/autoailabs/agentshield

🚫 Zero Telemetry

No analytics, no tracking, no usage reporting. We have no idea how you use it, and that's by design.

Verify: grep -r "telemetry\|analytics\|tracking" src/

🔑 No Credentials Needed

AgentShield never asks for your API keys or cloud credentials. It operates alongside your agents, not inside them.

Verify: no env vars required to run

📝 Built-in Audit

Every decision (allow/deny) is logged with full context: timestamp, agent, tool, layer results, trace ID.

Verify: cat ~/.agentshield/audit.log

🔒 Your Data, Your Control

All configuration, policies, and logs stay on your filesystem. Export, delete, or move them anytime.

Verify: ls ~/.agentshield/ — all data is here
GET STARTED

Three Steps. Three Minutes.

From zero to fully protected agents in under three minutes.

1

Install

One command. No dependencies. No configuration files.

# Install globally or use npx npx autoai-agentshield
2

Add to Your MCP Config

Add AgentShield as an MCP server in your agent's configuration.

{ "mcpServers": { "agentshield": { "command": "npx", "args": ["autoai-agentshield"] } } }
3

Done

AgentShield is now protecting your agents. Every tool call passes through 5 defense layers automatically.

✓ AgentShield active ✓ 5 defense layers enabled ✓ 60 threat patterns loaded ✓ Audit logging started ✓ Your agents are protected
PRODUCT POSTER

AgentShield at a Glance

View full poster ↗
INTERACTIVE DEMO

Try AgentShield Right Now

No installation needed. This runs entirely in your browser.

AgentShield Interactive Demo LIVE
Select a prompt and click "Scan with AgentShield" to see the result...
PRICING

Start Free. Scale When Ready.

No credit card required. Start free, unlock advanced detection and team features as you grow.

Free
£0
For individual developers
  • 20 threat patterns (of 60+)
  • Basic prompt injection detection
  • CLI scanning
  • Single agent monitoring
  • Advanced patterns (encoding, invisible chars)
  • Policy engine
  • Team dashboard
Get Started Free
Pro
£20/mo
For teams shipping AI products
  • All 60+ threat patterns
  • Advanced detection (encoding, unicode, delimiter)
  • Custom policy engine
  • Up to 25 agents
  • Slack/webhook alerts
  • Audit log export
  • Priority support
Start Pro Trial
Enterprise
£40/mo
For organisations at scale
  • Everything in Pro
  • Unlimited agents
  • SIEM integration (Splunk, Datadog)
  • Custom threat patterns
  • Team dashboard & RBAC
  • SOC 2 compliance reports
  • Dedicated support
Contact Sales
Custom
Custom
For regulated industries
  • Everything in Enterprise
  • On-premise deployment
  • Custom SLA
  • White-glove onboarding
Talk to Us

Never let an AI agent run
unprotected again.

Join hundreds of teams who ship AI agents with confidence. Start free, upgrade when you're ready.