How many agent tools actually work?
As of 2026-11-11 14:47 UTC, 68,896 of the 92,980 agent tools we index (74.1%) answered our latest probe, 44,063 (47.4%) passed a full protocol check, and 39,783 (42.8%) are ready for an agent to call.
As of 2026-11-11 14:47 UTC How we measure JSON All statistics
Answering is common. Passing a full check is not.
| Protocol | Indexed | Reachable | Verified | Ready (A/B) | Median latency | 100% uptime, 30 days |
|---|---|---|---|---|---|---|
| x402 APIs | 24,411 | 41.1% | 31.3% | 21.2% | 222 ms | 36.6% |
| MCP servers | 61,235 | 85.9% | 55.9% | 54.9% | 371 ms | 79.5% |
| A2A agents | 7,334 | 85.7% | 30.0% | 13.5% | 373 ms | — |
Shares are of all indexed listings of that protocol. Uptime counts listings with at least a week of daily probes.
How each listing's latest probe ended
One outcome per listing, from the HTTP status of our latest probe. A listing that asks for credentials is up, but an agent needs a key before it can use it.
x402 APIs
| Outcome | Listings | % |
|---|---|---|
| Answered the probe
|
9,998 | 41.0% |
| Answered, but asks for credentials (401/403)
|
206 | 0.8% |
| Endpoint not found (404)
|
3,211 | 13.2% |
| Server error (5xx)
|
6,850 | 28.1% |
| Other client error (4xx)
|
1,038 | 4.3% |
| No response (timeout, DNS or TLS failure)
|
3,074 | 12.6% |
| No callable endpoint
|
34 | 0.1% |
MCP servers
| Outcome | Listings | % |
|---|---|---|
| Answered the probe
|
36,777 | 60.1% |
| Answered, but asks for credentials (401/403)
|
16,205 | 26.5% |
| Endpoint not found (404)
|
2,583 | 4.2% |
| Server error (5xx)
|
1,534 | 2.5% |
| Other client error (4xx)
|
1,488 | 2.4% |
| No response (timeout, DNS or TLS failure)
|
2,025 | 3.3% |
| No callable endpoint
|
623 | 1.0% |
A2A agents
| Outcome | Listings | % |
|---|---|---|
| Agent Card fetched and valid
|
6,287 | 85.7% |
| Card fetched, with errors
|
591 | 8.1% |
| Card not fetched (no response or error)
|
456 | 6.2% |
Is it still up tomorrow?
Share of daily probes answered over the last 30 days, for listings we have probed for at least a week.
x402 APIs · 23,796 tracked
| Uptime | Listings | % |
|---|---|---|
| 100% |
8,700 | 36.6% |
| 95–99% |
218 | 0.9% |
| 80–94% |
587 | 2.5% |
| 50–79% |
572 | 2.4% |
| <50% |
13,719 | 57.7% |
MCP servers · 58,798 tracked
| Uptime | Listings | % |
|---|---|---|
| 100% |
46,745 | 79.5% |
| 95–99% |
2,518 | 4.3% |
| 80–94% |
1,433 | 2.4% |
| 50–79% |
686 | 1.2% |
| <50% |
7,416 | 12.6% |
A2A agents · 0 tracked
Not enough history yet.
If you build agents
- Check before you call. Every listing on this site carries its latest probe result, and the API and MCP server can filter to
health=ok. - Prefer readiness A or B. Only 39,783 listings reach it; they declare auth, schemas and pricing an agent can read.
- "Up" is not "usable": many MCP servers answer with 401/403 and need a key first. Their listing says which auth they declare.
- Paid x402 endpoints should answer 402 with a valid challenge. A 200, 404 or 5xx at the paid path means the listing is stale or misconfigured.
If you run a listing
- Your listing page shows where your service stands against reachable peers of the same protocol, and the criteria that cost the most points.
- Claim it to fix the description, add auth and pricing details, and get re-checked: List your service.
How we measure
- reachable
- answered our latest daily probe
- verified
- valid 402 challenge (x402), full initialize + tools/list (MCP), JSON-RPC reply (A2A)
- ready
- readiness grade A or B
Every listing is probed once a day from our servers. We never pay an x402 endpoint or call a paid tool; x402 checks stop at the 402 challenge. Full method: /stats#method and /api/v1/methodology.
Cite as: agent-tools.cloud, "How many agent tools actually work?", 2026-11-11, https://agent-tools.cloud/stats/usable. Data CC BY 4.0.