MASSAT - Security Audit for AI Agent Systems
Open-source security audit framework covering all 10 OWASP Agentic AI threat categories (ASI01-ASI10). Scan multi-agent systems for prompt injection, data exfiltration, broken access control, and 7 more categories. Returns risk scores (0-100) with remediation priorities. Free tier: 10 audits/day. Proven on a 94-agent production fleet. Complements Microsoft Agent Governance Toolkit.
Overview
MASSAT - Multi-Agent System Security Audit Toolkit
The open-source security audit framework for AI agent systems. Covers all 10 OWASP Agent Security Index (ASI) categories. Used in production to audit a 94-agent fleet.
Get a Free Audit in 30 Seconds
curl -X POST https://craigmbrown.com/api/audit \
-H "Content-Type: application/json" \
-d '{"repo": "https://github.com/your-org/your-agent-repo"}'
Returns JSON with risk score, findings by severity, and link to full HTML report.
What MASSAT Checks
| Category | OWASP ID | What It Catches |
|---|---|---|
| Unbounded Agency | ASI01 | Agents with no permission boundaries or tool restrictions |
| Unsafe Tool Use | ASI02 | Direct shell access, unvalidated file operations, SQL injection |
| Insecure Communication | ASI03 | Unencrypted inter-agent messaging, missing TLS |
| Memory Poisoning | ASI04 | RAG injection vectors, unvalidated memory writes |
| Inadequate Sandboxing | ASI05 | Code execution without isolation, container escapes |
| Excessive Permissions | ASI06 | Over-scoped API keys, admin privileges on read-only agents |
| Identity Spoofing | ASI07 | No agent authentication, missing delegation proofs |
| Weak Oversight | ASI08 | No human-in-the-loop for critical actions |
| Supply Chain | ASI09 | Unpinned dependencies, unverified model sources |
| Denial of Service | ASI10 | No rate limiting, unbounded resource consumption |
API Reference
POST /audit - Run Security Audit
# Audit a GitHub repo (free, 10/day)
curl -X POST https://craigmbrown.com/api/audit \
-H "Content-Type: application/json" \
-d '{"repo": "https://github.com/user/agent-repo"}'
# Audit with payment (unlimited, full scope)
curl -X POST https://craigmbrown.com/api/audit \
-H "Content-Type: application/json" \
-H "X-402-Payment: <ecash-token>" \
-d '{"repo": "https://github.com/user/agent-repo"}'
Response:
{
"audit_id": "audit-20260405-004858-f31d9003",
"risk_score": 4.3,
"risk_level": "medium",
"critical": 0,
"high": 0,
"medium": 6,
"low": 4,
"report_url": "https://craigmbrown.com/audits/audit-20260405-004858-f31d9003.html",
"get_passport": "https://craigmbrown.com/api/onboard?audit_id=audit-20260405-004858-f31d9003",
"subscribe": "https://craigmbrown.com/api/subscribe"
}
POST /subscribe - Get Security Updates
curl -X POST https://craigmbrown.com/api/subscribe \
-H "Content-Type: application/json" \
-d '{"email": "you@company.com", "name": "Your Name", "company": "Acme AI"}'
GET /audit/{id} - Retrieve Full Report
curl https://craigmbrown.com/api/audit/audit-20260405-004858-f31d9003
GET /health - Service Status
curl https://craigmbrown.com/api/audit/health
Blog Posts
- Security Auditing a 94-Agent Fleet: Before & After MASSAT - How we went from 4.3 to 4.0 risk score across 30 hardened agents
- 93% of AI Agents Have Zero Security - Why agent identity and audit infrastructure matters now
- OWASP ASI01-10 for AI Agent Builders - Practical guide to each threat category with code examples
Real Audit Examples
See examples/ for sanitized production audit reports from 5 different fleet types:
- SFA Fleet (19 Single File Agents) - Score: 4.3
- Orchestrator Fleet (13 coordination agents) - Score: 4.3
- Communication Fleet (6 WhatsApp/notification agents) - Score: 4.3
- Marketplace Fleet (25 BlindOracle DeFi agents) - Score: 4.3
- MCP Server (Context Oracle tool server) - Score: 4.5
Connection to BlindOracle
MASSAT is the security layer for the BlindOracle agent marketplace:
- Every marketplace agent must pass a MASSAT audit before activation
- Audit results are embedded in ERC-8004 agent passports
- Continuous auditing available for marketplace operators ($99/mo)
- Delegation proofs (15 kinds) are verified against MASSAT findings
Agent Onboarding Flow:
1. curl /api/audit -> Get security score
2. curl /api/onboard -> Get ERC-8004 passport (includes audit results)
3. curl /api/subscribe -> Join mailing list for security updates
4. Marketplace active -> Agent listed on BlindOracle with verified badge
Repository Structure
massat-framework/
README.md # This file
LICENSE # MIT
api/
openapi.yaml # OpenAPI spec for the audit API
blog/
security-auditing-94-agent-fleet.md
93-percent-zero-security.md
owasp-asi-guide.md
examples/
audit-reports/ # Sanitized production audit JSON
curl/ # Ready-to-run curl examples
docs/
getting-started.md # Quick start guide
api-reference.md # Full API docs
threat-model.md # OWASP ASI01-10 detailed threat model
assets/
massat-badge.svg # Badge for GitHub READMEs
Links
- Live API: craigmbrown.com/api/audit
- BlindOracle Marketplace: craigmbrown.com/blindoracle
- Agent Passports: blindoracle-docs
- Whitepaper: Security Auditing a 94-Agent Fleet
License
MIT - See LICENSE for details.
Built by Craig Brown | Powered by BlindOracle
Server Config
{
"mcpServers": {
"massat": {
"command": "python",
"args": [
"-m",
"massat.server"
],
"env": {
"MASSAT_API_URL": "https://craigmbrown.com/api/v1/massat"
}
}
}
}Project Info
Created At
2 months agoUpdated At
2 months agoAuthor Name
craigmbrownStar
-Language
-License
-Recommend Servers
View AllSerper MCP Server
@garymengcom
A Serper MCP Server
Python
a year ago
Playwright Mcp
@microsoft
Playwright MCP server
TypeScript
10 months ago
Juhedeepsearch
9 hours ago
Filesystem
@modelcontextprotocol
2 months ago
Memory
@modelcontextprotocol
a year ago
20 0 Game Mcp Server
@rocnubie
11 hours ago