MCP security scanning and detection.
Scan your MCP servers for vulnerabilities, deploy tripwires that detect prompt injection, and monitor everything from one dashboard.
Get started freeNo credit card required.Scan. Protect. Monitor.
Find what's dangerous, deploy tripwires, and watch everything from one dashboard.
Each step takes one command.
Find what's dangerous
$ npx decoy-scanScans every MCP server on your machine. Tool risk, prompt injection, transport security, OWASP mapping. No signup, no token, no config.
Add to CI/CD
$ uses: decoy-run/decoy-scan@v1GitHub Action that scans on every PR. Fails the build on critical issues or poisoned tool descriptions.
Real results
We scanned modelcontextprotocol/servers (82k stars).
Critical
High
Poisoned
The official MCP reference servers. Maintained by Anthropic. Read the full report.
Features
Security Scanning
50+ checks across tool risk, input validation, transport security, and OWASP mapping. Works with 8 MCP clients. Zero dependencies.
Tripwire Detection
12+ decoy tools that blend in with your real MCP servers. No honest agent calls them — but a compromised one will. Near-zero false positives by design.
See what your agents have access to
One command scans every MCP server on your machine. Tool risk, input validation, transport security, permission scope, and OWASP mapping. No account required.
What the scan covers
The scan command finds every installed MCP server, inventories each tool, and runs a full security analysis — risk classification, input validation, SSE transport, and permission scoring.
- Tool Risk — classify every tool from low to critical
- Input Validation — unconstrained params, missing schemas
- Transport Security — SSE auth, TLS, CORS, rate limits
- Permission Scope — over-privileged servers, dangerous combos
npx decoy-scanNo signup, no token, no config.
uses: decoy-run/decoy-scan@v1GitHub Action — scans on every PR, fails on critical issues.
Decoy Tripwires
Deploy realistic decoy tools that blend in with your real MCP servers. 12 built-in tripwires plus dynamically generated tools unique to your deployment. No honest agent would call them — but a compromised one will.
- Dynamic tool generation — unique tripwires per deployment
- Looks real to any AI model
- Near-zero false positives by design
- Deploys in under 30 seconds
Agent Identity & Telemetry
Every agent that connects gets automatically fingerprinted with rich session telemetry. Track client identity, tool call sequences, and behavioral patterns in real time.
- Automatic agent fingerprinting
- Session telemetry (client ID, tool call sequences)
- Per-agent trigger history
- Status tracking (active / paused / revoked)
Open Source Tools
Start with free, open-source tools that work standalone. Decoy Scan finds vulnerabilities. Decoy Tripwire detects prompt injection. Decoy Guard is the paid platform that ties it all together.
- decoy-scan — MCP vulnerability scanner (npm, zero dependencies)
- decoy-tripwire — tripwire detection (npm, zero dependencies)
- Decoy Guard — dashboard, threat intel, security testing (paid)
Your agent, aware of its own security
Connect your AI agent to Decoy Guard via MCP. It gets direct access to trigger logs, scan results, threat intel, and red team assessments — no tab-switching required.
Guard MCP Server
One line to install. 9 tools your agent can call — check status, view triggers, run scans, pull threat intel, and launch red team assessments. All tier-gated and rate-limited.
claude mcp add decoy-guard --transport http "https://app.decoy.run/guard/TOKEN"Works with Claude Code, Claude Desktop, and any MCP client.
- Stateless HTTP — no SSE, no long-lived connections
- Tier-gated tools — Free sees 5, Pro sees all 9
- AI-powered red teaming via Claude API
- 60 req/min rate limiting, 256KB body limit
decoy_statusdecoy_triggersdecoy_agentsdecoy_scan_rundecoy_riskdecoy_feeddecoy_redteamYour MCP servers expose tools to AI agents — tools that can read files, execute commands, access credentials. If any are misconfigured, over-permissioned, or poisoned, your agent will use them exactly as configured. Most teams don't find out until something breaks.
Decoy finds these problems before they matter. One scan command, 50+ security checks, every MCP server on your machine. No agents involved. No account required.
For teams that ship agents to production, we add continuous monitoring. Tripwires sit alongside your real tools — realistic decoy tools that no honest agent would ever call. When a prompt injection triggers one, you find out immediately.
Not after the fact. Right then.
Real-time trigger monitoring
Track triggers, identify patterns, and understand what your agents are doing — as it happens.
Threat Map
See where triggers come from. Trigger events across your agent infrastructure, mapped geographically.
Trigger Analytics
Understand trigger patterns. Analyze by severity, category, and agent to see what's happening across your deployment.
Real-Time Detection
Triggers fire the moment an agent calls a decoy tool.
Sub-Millisecond Response
No scanning, no delays. Tripwires respond in under a millisecond.
Daily Digests
Get a daily summary of all trigger events across your agents.
Threat Intelligence
Track emerging patterns across the MCP ecosystem.
Works with every MCP host
One install protects all your AI agents. Decoy auto-detects and configures every MCP-compatible host on your machine.
Built for Production
Edge-deployed globally with full trigger logging and real-time agent monitoring.
Scale With Confidence
Thousands of concurrent agent connections, zero performance impact. Cloudflare's global edge network.
- Multi-token management
- Per-agent access controls
- Webhook & API integrations
Auto-Expiring Data
Trigger data auto-deletes after 90 days.
Edge Deployment
Runs globally with low overhead.
Full Trigger Logging
Agent, tool, args, time, and severity.
Flexible Alert Routing
Route to Slack, email or webhook.
Get started in under a minute
Scan your servers, deploy tripwires, and start monitoring. No credit card required.
Get started free