greyridge-x402-tools

AFG: agents hire agents (test money) MCP server — security & directory-readiness scorecard

Server: https://afg.ai/mcp Scanned: 2026-10-22 · Tools advertised: 18 · Injection verdict: clear · Directory: 18 Claude Directory blocker(s)

Sandbox marketplace where AI agents hire agents for verified outcomes. Test money only.

Before an agent — or a user — connects it, every tool definition it advertises is read by the model, so a poisoned description or a hidden instruction can steer the agent before any tool runs. This scorecard is an independent, static read of that advertised metadata, run against the AFG: agents hire agents (test money) server on the date above.

What the scan found

• No tool-poisoning or prompt-injection patterns were found across the 18 tool definition(s) this server advertises.
• Claude Connector Directory readiness: 18 blocker(s). This server advertises 18 tool definition(s) that would likely fail Anthropic’s structural review criteria. The Connector Directory review runs about 20 days, so a rejection is expensive — the itemized list of which criteria fail, and how to fix each, is one paid call away.

The scan is static analysis of the advertised tools/list metadata only — it never invokes a tool and never sends data anywhere. It is a point-in-time read: an MCP server’s tool definitions can change at any time (that is the “drift” risk this scorecard exists to make visible), so a clean result today is not a guarantee for tomorrow.

Re-run this scan yourself

No account, no key:

curl "https://x402-data-api.sigrunner.workers.dev/scan/mcp/preview?url=https://afg.ai/mcp&directory=true"

That free response gives you the verdict and the directory blocker count. For the itemized findings — which tools, which OWASP LLM01/LLM08 pattern, and for the directory check which criteria fail and how to fix each — the zero-egress paid report is one call:

curl -s -X POST https://x402-data-api.sigrunner.workers.dev/scan/mcp -H 'content-type: application/json' \
  -d '{"directory":true,"url":"https://afg.ai/mcp"}'

$0.05 USDC via x402 on Base, no account. To keep a server honest as it changes, continuous drift monitoring with alerts is $9.99/mo — see pricing.

More

• Scan any MCP server for prompt injection / tool poisoning
• Why was my Claude connector rejected? — directory readiness criteria
• All scanned MCP servers

Independent scan by Grey Ridge Signals. Measured on 2026-10-22; re-run the command above for a current result. Questions: contact@greyridgesignals.ai

This site is open source. Improve this page.