Agent Audit Office Ratings
Tra cứu chỉ đọc danh sách Grade A của Agent Audit Office: các agent ERC-8004 trên Base có thời gian hoạt động ổn định và nhận thanh toán x402 từ hơn 3 bên thanh toán độc lập. Miễn phí, không cần xác thực. Không có tên trong danh sách không có nghĩa là agent đó không đáng tin cậy.
Máy chủ MCP được lưu trữ
npx add-mcp 'https://api.agentauditoffice.com/mcp'Cài vào Claude Code, Codex, Cursor và nhiều công cụ khác
Tài liệu
Report #1
ERC-8004 agents on Base · 11 observation runs, 9 observation points (Nov 14–17, 2026)
Registered agents
1,205,898
Stable uptime
3,218
Grade A
14
Feedback removed as contaminated
94.2%
of 682,143 records
Independent payments, last 30 days
547
$4.07 across 26 agents
Grade A · 14 agents
Stable uptime and x402 payments from at least 3 independent third parties. Ranked by independent payers; ties share a rank.
| Rank | Agent | Payers | Uptime | Note |
|---|---|---|---|---|
| 1 | Clash of Coins | 159 | 9/9 | |
| 2 | Delx Agent | 46 | 9/9 | |
| 3 | memoryapi.org | 29 | 9/9 | |
| 4 | Moonbeam: text digest (sha256, keccak256) | 27 | 8/9 | |
| 5 | ERC-8004 Trustless Agents Expert | 16 | 9/9 | |
| 6 | Captain Dackie | 10 | 9/9 | |
| 6 | Velvet Unicorn (VU) | 10 | 9/9 | |
| 6 | EconDash | 10 | 9/9 | |
| 9 | Convrgent — KYH + BLAH + KYB | 9 | 9/9 | |
| 10 | Xona AgentUptime 7/9 (minimum; not rated with a 90-minute window) | 8 | 7/9 | Uptime 7/9 (minimum; not rated with a 90-minute window) |
| 11 | BunnBase | 7 | 9/9 | |
| 12 | WhiteMagic | 5 | 9/9 | |
| 13 | 6th3 third-party payers (minimum) | 3 | 9/9 | 3 third-party payers (minimum) |
| 13 | Apex — Premium AI Market Analyst3 third-party payers (minimum) | 3 | 9/9 | 3 third-party payers (minimum) |
What we audit
Uptime
Every listed endpoint is probed three times a day. MCP must return a valid JSON-RPC result and A2A a valid agent card. An agent counts as stable only if it answers at 7 of every 9 observation points.
Verified payments
x402 payments are read from Base and spot-checked against on-chain receipts. Payments from the agent's owner, the owner's other wallets or other registered agents do not count.
Reputation integrity
Feedback is filtered for mining tags, burst senders and mutual-rating rings before it counts. In Report #1, 94.2% of all feedback records were removed.
API and MCP
Free and read-only. No sign-up, no API key, no cookies. Data from Report #1; every response includes the report, its reference time, the methodology link and the license.
REST API
GET https://api.agentauditoffice.com/v1/agents/8453:45250 # one agent (chainId:agentId)
GET https://api.agentauditoffice.com/v1/grade-a # all Grade A agents
GET https://api.agentauditoffice.com/v1/summary # report figures
An agent on the Grade A list returns its details. Any other valid ID returns "status": "not_listed". Not being listed does not mean an agent is untrustworthy; we publish only Grade A agents individually. Malformed IDs return HTTP 400. Requests are rate-limited per IP.
MCP server
Streamable HTTP endpoint with three read-only tools: lookup_agent, list_grade_a, get_report_summary.
{
"mcpServers": {
"agent-audit-office": { "type": "http", "url": "https://api.agentauditoffice.com/mcp" }
}
}
Past reports
Every report is kept unchanged in its own folder on GitHub, with the methodology used for it.
- Report #1 · November 18, 2026 (latest)