MarketNow now maps 100% to the OWASP MCP Security Cheat Sheet

#mcp #ai #security #owasp

OWASP released a MCP Security Cheat Sheet. Here is how MarketNow aligns.

The OWASP MCP Security Cheat Sheet defines 12 security controls for MCP servers. MarketNow now maps 100% to all 12.

4 are live today. 8 are in the roadmap (v5.1-v6.0).

The 12 OWASP controls

#	OWASP Control	MarketNow Status
1	Verify tool descriptions haven't changed	v5.1 — Tool Fingerprinting API (SHA-256 hashes)
2	Validate input/output schemas	v5.1 — Schema hash in fingerprint
3	Monitor for tool poisoning	v5.0 partial (Interceptor) → v5.2 (full drift detection)
4	Implement least privilege	v5.3 — Capability Graph (filesystem, network, shell, credentials)
5	Log all tool invocations	v5.3 — Agent Identity + Task Identity
6	Isolate tool execution	✅ LIVE — L2.5 gVisor sandbox
7	Scan for prompt injection	✅ LIVE — L1.9 (32 rules, 10 categories)
8	Monitor runtime behavior	v5.2 — Behavioral Baseline + Drift
9	Verify supply chain integrity	v5.1 — Provenance / SLSA
10	Implement revocation	✅ LIVE — ATC revocation (Ed25519, requires ca_secret)
11	Detect typosquatting	v6.0 — Levenshtein distance
12	Track data flow	v5.4 — Data Flow Tracking + Trajectory Security

Try the compliance API

curl https://marketnow.site/api/owasp?compliance

Returns the full checklist with MarketNow implementation details for each control.

Tool Fingerprinting (OWASP control #1)

Generate a SHA-256 fingerprint of any MCP tool:

curl https://marketnow.site/api/owasp?fingerprint=mn-gen-00003

{
  "fingerprint": {
    "tool_hash": "12a893c727a9e8be1e78410719fbec16...",
    "name_hash": "...",
    "description_hash": "...",
    "algorithm": "SHA-256"
  },
  "capabilities": {
    "filesystem": "none",
    "network": "none",
    "shell_execute": false,
    "credential_access": false
  }
}

For full tools/list fingerprinting (hashes every tool definition individually):

curl -X POST https://marketnow.site/api/owasp \
  -H "Content-Type: application/json" \
  -d x27{"action":"fingerprint","skill_id":"mn-xxx","tools_list_response":{"tools":[{"name":"read_file","description":"...","inputSchema":{"type":"object"}}]}}x27

Returns per-tool hashes: name_hash, schema_hash, description_hash, and an overall tools_hash.

If any hash changes at runtime, the Trust Card should be auto-revoked. That is tool poisoning detection.

Capability Manifest (OWASP control #4)

curl https://marketnow.site/api/owasp?capabilities=mn-gen-00003

{
  "capability_manifest": {
    "filesystem": "read-only",
    "network": "outbound",
    "shell_execute": false,
    "process_spawn": false,
    "credential_access": false,
    "database": false,
    "prompt_handling": true,
    "risk_factors": {
      "high_risk": false,
      "network_exfil": false,
      "data_access": true
    }
  }
}

Organizations can define policies: "My agents can only use tools with score >= 8 AND no filesystem AND no shell." (Coming in v5.3)

The numbers

1,211,488 security checks performed
1,030 threats detected
80 skills quarantined (critical — malware, hacking tools)
8,288 verified safe (score >= 8)
57 Ed25519-signed Trust Cards
9,248 MCP skills analyzed
12 OWASP controls mapped (4 live, 8 roadmap)
100% OWASP MCP Cheat Sheet coverage

Roadmap

v5.1 VERIFICATION — Full fingerprints, provenance, confidence scoring, transparency log
v5.2 BEHAVIOR — Behavioral baselines, drift detection, network/fs/process analysis
v5.3 POLICY — Capability graph, org policies, agent identity, approval workflows
v5.4 TRAJECTORY — Multi-tool attack chains, cross-tool escalation, data flow tracking
v6.0 PLATFORM — Multi-protocol (MCP+A2A+OpenAI), AgentBOM, cross-agent trust

Full roadmap: https://github.com/edgarfloresguerra2011-a11y/marketnow/blob/master/ROADMAP.md