# mcp-security-suite

**One enterprise MCP security brand** over the previously separate sudo-ai-git
deterministic no-LLM servers. For the "Agent Identity / Enterprise-Ready
Security" gap in MCP platforms (e.g. kubernetes-sigs/mcp-lifecycle-operator).

## What it unifies (12 tools, 4 domains)

| Domain | Prefix | Server | Tools | What it does |
|---|---|---|---|---|
| verify-claim | `vc_` | mcp-verify-claim | tier_claim, check_claim, report_templates, tier_definitions | Evidence-gated tiered claim reporting (FACT/INFERENCE/SPECULATION) — never fabricates |
| skill-sec | `sec_` | mcp-skill-sec | audit_text, audit_skill_file, audit_directory, rule_list | Audits agent skills/prompts vs 8 supply-chain attack patterns |
| token-saver | `tok_` | mcp-token-saver | profile_tokens, duplicate_payload_report | Deterministic token profiler (per-role spend, dedupe, savings) |
| auth-audit | `auth_` | mcp-auth-audit | audit_server_auth, audit_token_mode | MCP server auth-mode audit |

`mcp-schema-lint` (registry-manifest + tools/list validator) ships as companion CLI.

## Run

```bash
python -m mcp_security_suite        # stdio MCP server, all 12 tools
```

## Validation (P2 gate)

`validate_suite.py` imports each component server and asserts the consolidated
`server.json` lists exactly the real, loadable 12 tools — no missing, no
phantoms. `RESULT: PASS`. End-to-end tool execution proven (e.g. `sec_audit_text`
returns `verdict: FLAG` on a curl-pipe-sh payload).

## Honest scope

This is a **brand/registry unification** — the tool implementations live (and are
tested) in their component repos; this package re-exports them under one
entrypoint + registry manifest. It does NOT fork implementation code. All
backends are deterministic, no-LLM, no-network.