How many MCP servers actually answer a tools/list call? What does an x402 API usually charge? How many A2A agents reply when you send them a message? Until now you could only guess, or probe everything yourself. Today agent-tools.cloud publishes the answers every day at /stats.
The numbers come from our own probes of every listing in the directory, not from what each service says about itself. Here is what is new and how to use it.
Three things the data already shows
The snapshot from 12 November 2026 covers 61,862 MCP servers, 24,531 x402 APIs and 7,372 A2A agents.
First, being listed is not the same as being usable. 85.7% of MCP servers are reachable and 58.1% complete a full handshake. For x402 the figures are 41.1% reachable and 31.3% returning a valid payment challenge. A2A agents load their Agent Card 85.6% of the time, yet only 29.7% answer a JSON-RPC call.
Second, when a call fails, it is rarely because the server is slow. The median successful probe returns in 205 ms for x402, 326 ms for MCP and 398 ms for A2A, and over 99% of reachable endpoints answer within 2.5 seconds. Failures come from servers that are gone: among x402 listings, 28.0% return a server error, 13.2% a 404 and 12.5% nothing at all.
Third, paid tools are cheap. Across 7,148 x402 APIs that publish a price, the median call costs $0.01 and 65% cost a cent or less. Nine in ten settle on Base.
What you can open today
/stats is the daily overview: how many listings are indexed, reachable, verified and ready, with trend lines as history builds up. It also shows why probes fail, which protocol versions are in use, how readiness and latency are distributed, prices, payment chains, adoption signals and what people search for in the directory.
/stats/x402, /stats/mcp and /stats/a2a show the same sections for one protocol at a time.
/stats/usable answers one question per protocol: of everything listed, how much could an agent call right now? Each step from indexed to ready is a bar, so you can see where the drop happens.
Topic pages break the directory down by subject. /stats/topics/payment, for example, shows how many payment tools exist in each protocol, how many are reachable and ready, their median readiness and latency, and the most ready ones. Every topic is linked from the overview.
Monthly reports summarise each month on one page. November 2026 is the first and fills in as the month goes on.
Every listing page now has a peer section. It places the listing against all reachable listings of the same protocol on readiness and latency, and for each gap it names the check that failed and what fixes it. If you run a service, this is the quickest way to see what is holding it back.
Built for agents as well as people
Everything on the stats pages is available as data, under CC BY 4.0:
- /stats.json for the full snapshot, /stats/daily.csv for headline numbers by day
.jsonversions of the usable, topic and report pages- the
directory_statstool on our MCP server, withview,topicandmontharguments - the
directory_statsskill on our A2A agent
Without an account you see the top three listings per protocol on each topic and the top ten search terms. With a free key, you get the full lists: ten listings per protocol per topic, the top 100 search terms and the complete daily history. Agents can get a key with one call to POST /api/v1/keys and send it as Authorization: Bearer <key>; people can sign in with GitHub. Then add ?full=1 to any of the data URLs above.
Search terms, kept private
The search section counts what people and agents look for, so tool builders can see where demand is. Only aggregate counts per day are stored, a term needs at least five searches before it is shown, and anything that looks like an email, URL, key, address or long identifier is dropped before it is counted. Details are in our privacy notes.
Tell us what is missing
The snapshot runs once a day after the probe pass. How every number is computed is on /about#statistics and in /api/v1/methodology. If you build or operate agent tools and want a figure we do not show yet, or think one of ours is wrong, claim your listing or write to [email protected].