Comparison
AgentGate vs Cisco skill-scanner + mcp-scanner
Cisco splits agent security across two tools:skill-scannerfor agent skills andmcp-scannerfor MCP servers — and its strongest engines need external API keys. AgentGate covers the whole agent config surface in one deterministic, key-free CLI, then locks and CI-gates it. Sources and verification dates indocs/COMPARISON.md.
| Capability | AgentGate | Cisco skill-scanner + mcp-scanner |
|---|---|---|
| Tools needed for full coverage | One CLI — MCP configs, skills, commands, plugins, lockfile, dependencies | Two separate tools: skill-scanner (skills) + mcp-scanner (MCP servers) |
| API keys required | None | Two of three mcp-scanner engines need keys (LLM provider / Cisco AI Defense) |
| Deterministic results | Yes — rule-based, reproducible in CI | LLM-as-a-judge engine is non-deterministic |
| Lockfile of the approved tool surface | agentgate.lock — names, descriptions, input schemas | None |
| CI drift gate | agentgate ci — non-zero exit + readable diff + SARIF | None |
| Advisory database | Public, structured, 110+ entries, free JSON API | None (pip-audit for Python deps, VirusTotal for binaries) |
| Client config discovery | Auto-discovers Claude, Cursor, VS Code, Codex, OpenCode, and 20+ more | Config file input (mcp-scanner) |
| Runtime / language | TypeScript, Node 22 — one npx away for JS teams | Python 3.11+ |
Where AgentGate wins
- One tool, one report, one CI step for configs, skills, commands, plugins, lockfile drift, and dependencies.
- No API keys: results are deterministic and reproducible — the same input always yields the same findings.
- Lockfile + drift gate defends against rug-pulls that any point-in-time scanner misses.
- Public advisory database cross-checked automatically on every scan.
Where Cisco wins (honest)
- LLM-as-a-judge can catch novel semantic attacks that rule-based analysis has no signature for.
- YARA engine, VirusTotal binary lookups, and PyPI sandbox scanning cover angles AgentGate does not attempt.
- Backed by Cisco AI Defense research with 3,400+ combined GitHub stars (2026-08).