# Seller QA and Security MCP Suite

Thirteen tools across four audited MCP servers, plus four Agent Skills and the full security package that most listings promise and never publish.

| | |
| --- | --- |
| Version | 1.0.0 |
| Price | 149 USD |
| Licence | Single-seat |
| Entry point | `README.md` |
| Source | Bundle of security_posture, listing_auditor, payment_verifier and seller_qa |

## What it does

Four read-only MCP servers with four Agent Skills and a published security package: per-server audit reports, permissions matrix, and published exclusions.

## Tools

| Tool | Description |
| --- | --- |
| `scan_secrets` | Scan a directory tree for API keys, private keys and credentials. Returns redacted findings. | | `check_permissions` | Compare declared role permissions against a least-privilege policy. | | `ssrf_audit` | Audit URLs against a policy that denies private, loopback and link-local targets. | | `read_only_report` | Report the read-only posture of a directory and an optional permissions manifest. | | `score_listing` | Score a listing against the 120-point rubric and return every lost point. | | `seo_validate` | Validate title, description, tags and images for search readiness. | | `improve_listing` | Return the fix list for a listing, biggest recoverable points first. | | `verify_payment` | Confirm recipient, amount, status and timestamp for one transaction. | | `recent_orders` | Summarise recent transactions for a wallet. | | `wallet_status` | Report wallet address, network and read-only posture. Never signs. | | `preflight_publish` | Dry-run a publish against the marketplace gate rules. Writes nothing and calls no platform. | | `reconcile_listing` | Compare local listing records against a platform report. | | `verify_expectation` | Verify an expected value against an observed one, with optional numeric tolerance. |

## Network

Allow-listed gateways only: `api.trongrid.io`, `tronscan.org`, `bsc-dataseed.binance.org`.

## Security

- Read-only: no file is written and no process is started.
- Findings are redacted before they are returned.
- Paths are confined to the root you pass.

## Requirements

- Python 3.10 or above
- No account, no API key, no telemetry

## Agent Skills

| Skill | Purpose |
| --- | --- |
| `pre-publish-security-gate` | Run the leak and permission checks before anything is published. | | `listing-audit-triage` | Turn a listing score into an ordered fix list. | | `payment-verification-triage` | Confirm a USDT payment before delivering an order. | | `pre-publish-gate` | Dry-run the marketplace gate and see every blocker first. |

## Tests

```bash
python -m pytest -q # suite tests for this server
```

Every security claim above names a test that exists:

- `test_ssrf_blocks_private_network`
- `test_no_subprocess_spawn`
- `test_path_traversal_blocked`
- `test_dry_run_isolation`

## Licence and exclusions

- Single-seat licence unless a team licence is stated above. - No shell execution, no browser or account automation, no key custody. - The exclusion list is published in `security/EXCLUSIONS.md` in the team bundle so a reviewer can check it independently.