Comparison
mcpprobe and Cisco mcp-scanner: many analysers versus live conformance
Cisco's mcp-scanner is an open-source toolkit of analysers: YARA rules, an LLM judge, static code analysis and more, with an optional Cisco AI Defense API. mcpprobe focuses on what a running server actually answers. Here is what the public record says about each.
Facts about Cisco mcp-scanner checked on from the sources listed at the bottom. Corrections: hello@mcpprobe.tech.
Side by side
| Aspect | Cisco mcp-scanner | mcpprobe |
|---|---|---|
| Licence and size | Apache-2.0; about 1.1k GitHub stars (27 Oct 2026) | Apache-2.0 from v0.2 (v0.1.0 MIT), first release 27 Oct 2026 |
| Install | uv tool install --python 3.13 cisco-ai-mcp-scanner | pipx run --spec mcpprobe-scanner mcpprobe <url> |
| Analysers | YARA rules, LLM analysis via LiteLLM, Cisco AI Defense API (subscription), behavioural static code analysis, VirusTotal hash lookups, pip-audit, 20 readiness heuristics, a regex prompt-defence analyser | Deterministic protocol and authorization probes, description heuristics with an optional judge hook; scoped injection canaries and sandboxed stdio runs are coming soon |
| Targets | Remote SSE and streamable-HTTP servers, stdio commands, config files, static JSON snapshots, PyPI and npm packages (never executed) | Remote Streamable HTTP endpoints and local stdio servers you run |
| MCP authorization conformance | Not listed among its analysers | AUTH-01 to AUTH-04: no auth, RFC 9728 metadata, PKCE S256 and HTTPS, wrong-audience and query-string tokens |
| Independent false-positive data | One audit (Apr to Jun 2026): 27 detections across 33 servers, about 78% judged false positives; all 8 prompt-injection detections were false positives; 92.6% rated HIGH | None yet. Per-check rates will be published on the checks page once each check has 100 samples |
| Where your data goes | Local by default; the LLM analyser needs your model credentials and the AI Defense analyser sends data to Cisco | CLI sends no telemetry and makes no LLM calls; the judge is a hook for your own model |
Where Cisco mcp-scanner is the better choice
- You want to analyse source code statically, for example matching a tool's docstring against what its implementation does.
- You want YARA rules you can extend, VirusTotal lookups, or an integration with Cisco AI Defense.
- You need to scan packages without ever executing them.
Where mcpprobe is the better choice
- You need to know whether a deployed server enforces the MCP authorization spec and validates Origin.
- You want each finding proven by a request and a response rather than inferred from text.
- You want severity that is not inflated: mcpprobe never marks a finding Critical on heuristics alone.
- You want rug-pull diffs against a stored baseline and SARIF that drops straight into code scanning.
The short answer
Cisco mcp-scanner is a broad static and heuristic toolkit; mcpprobe is a dynamic tester for servers you operate. If you can read the code and want many lenses on it, start with Cisco's scanner. If you need to prove what the running server does, and want fewer, reproducible findings, use mcpprobe. mcpprobe's own false-positive rates are not measured yet, and this page will link them when they are.
Sources
- cisco-ai-defense/mcp-scanner repository (fetched 27 Oct 2026)
- AppSec Santa: MCP server security audit 2026 (Apr–Jun 2026)
- Decryption Digest: MCP server security scanners compared (31 Jul 2026)