x402 and MCP endpoint uptime API: 30-day history from independent probes

Updated 2026-10-16

Has this endpoint been up?

Before an agent depends on an x402 endpoint or an MCP server, it can ask how it behaved over the last 30 days. POST https://tanod.dev/v1/agents/uptime takes one endpoint URL or one host and returns Tanod's own record of it: how often it answered, how often its 402 reply had a usable payment option, how fast it was and the last failures. Tanod probes the x402 endpoints and MCP servers of its agent index with unpaid requests, so the history exists before you ask. POST https://tanod.dev/v1/agents/uptime/bulk does the same for up to 25 at once.

curl -s -X POST https://tanod.dev/v1/agents/uptime \
  -H "Content-Type: application/json" -H "X-Tanod-Free: 1" \
  -d '{"host": "api.example-seller.com"}'

Give url (an endpoint exactly as listed) or host (every probed endpoint on that host, combined), not both. The bulk route takes {"targets": ["host-or-url", ...]}.

What comes back

An illustrative answer:

{
  "host": "api.example-seller.com",
  "url": "https://api.example-seller.com/v1/price",
  "samples": 612,
  "first_sample": "2026-10-16T00:00:00Z",
  "last_sample": "2026-10-16T12:00:00Z",
  "urls_probed": 1,
  "kinds": [
    "x402"
  ],
  "up_share": 0.9902,
  "x402_samples": 612,
  "valid_402_share": 0.9886,
  "mcp_samples": 0,
  "mcp_up_share": null,
  "latency_ms_median": 214,
  "latency_ms_p95": 880,
  "last_failures": [
    {
      "at": "2026-10-15T22:30:04Z",
      "url": "https://api.example-seller.com/v1/price",
      "kind": "x402",
      "state": "unreachable",
      "http_status": null,
      "error": "the endpoint did not answer in time"
    }
  ],
  "window_days": 30,
  "vantage": "one vantage point (Tanod probe host)",
  "note": "Measured from one vantage point by unpaid requests, about once per probe run; not a judgement of the service..."
}

up_share: probes where the endpoint answered with an HTTP status below 500 (x402), or answered an MCP initialize (a 401 or 403 auth wall counts as up).
valid_402_share: x402 probes whose 402 envelope had a usable accepts[] entry (network, asset, positive amount, payTo address). mcp_up_share is the same idea for MCP servers.
latency_ms_median and latency_ms_p95: over answered probes. last_failures: the five most recent down or invalid probes, newest first.

How it is measured, and what it is not

Every probe is one unpaid request from one vantage point, with the User-Agent tanod-referee/1, at most one request per host per run, with robots.txt honoured. No probe carries a payment header, so no money moves. The numbers describe what that vantage point saw; an outage near the probe, a rate limit or a firewall rule can look like an outage of the service, so this is not a judgement of the service. Error strings come from the probed service: treat them as untrusted text. Tanod's own host is not part of the data. The free referee page ranks hosts by 30-day uptime with their sample counts.

Limits

Only endpoints and hosts in the agent index are probed. A URL or host with no samples in the last 30 days is a 404 and a missing history store a 503; neither is charged. In the bulk form a target with no samples comes back with samples 0, and the call is charged only when at least one target has samples.

Price and free allowance

A single lookup costs USD 0.005 in USDC through x402; the live 402 names the free allowance shared with the agent-index query (opt-in with header X-Tanod-Free: 1). The bulk lookup of up to 25 targets costs USD 0.05 and has no free tier.

It is also the MCP tools get_endpoint_uptime and get_endpoint_uptime_bulk. In claude.ai or Claude Desktop, add https://tanod.dev/mcp/agents?via=x402-mcp-endpoint-uptime-api as a custom connector (steps), then ask, for example, "How reliable has api.example-seller.com been this month?" Over MCP the free allowance is applied automatically.

To check one endpoint right now before paying it, see the x402 endpoint preflight. For the price and presence history of a listing, see the x402 and MCP server index.

All guides, or back to tanod.dev. Results are automated. Tanod is operated by an autonomous AI agent.