A week ago I counted every server in the official MCP Registry: 59,603 entries. Since then a small job has taken a snapshot every morning. Nine snapshots later, the registry lists 64,972 servers.
Disclosure: I build data tools for AI agents as Fast Data on Apify, and some of them are listed in this registry. The numbers below come only from the registry's public API.
| Snapshot (morning, UTC) | Servers (latest versions) | With a remote endpoint | Share remote |
|---|---|---|---|
| 8 Nov | 59,612 | 39,796 | 66.8 % |
| 10 Nov | 60,850 | 40,875 | 67.2 % |
| 12 Nov | 62,372 | 42,044 | 67.4 % |
| 14 Nov | 63,765 | 43,080 | 67.6 % |
| 16 Nov | 64,972 | 44,052 | 67.8 % |
That is +5,360 servers in eight days (+9.0 %), about 670 per day. The share of servers that offer a remote (hosted) endpoint instead of only a local package went up every single day, from 66.8 % to 67.8 %. New servers are more likely to be remote than the existing stock.
Grouping every listed server by the date its name was first published gives the chart above: 20,136 new servers between 17 October and 15 November, an average of 671 per day.
| Pattern | Number |
|---|---|
| Median new servers on a weekday | 667 |
| Median on a weekend day | 563 |
| Busiest day | Sunday 8 Nov: 1,311 |
| Next two | Mon 9 Nov: 965, Tue 10 Nov: 853 |
The weekend dip is what you would expect from people publishing by hand. The Sunday spike is not. A single day with twice the normal volume on a weekend looks like batch publishing, one organisation or tool pushing many servers at once. I have not checked which namespaces drove it; that is the next thing to look at.
The registry API is public, paginated with a cursor, and returns metadata per server version. The official metadata block tells you which version is the latest and when it was first published:
import collections, requests
URL = "https://registry.modelcontextprotocol.io/v0.1/servers"
first_seen, latest, remote, cursor = {}, 0, 0, None
while True:
params = {"limit": 100, **({"cursor": cursor} if cursor else {})}
page = requests.get(URL, params=params, timeout=60).json()
for s in page.get("servers", []):
meta = s.get("_meta", {}).get("io.modelcontextprotocol.registry/official", {})
name = s["server"]["name"]
first_seen[name] = min(first_seen.get(name, "9999"), meta.get("publishedAt") or "9999")
if meta.get("isLatest"):
latest += 1
remote += 1 if s["server"].get("remotes") else 0
cursor = page.get("metadata", {}).get("nextCursor")
if not cursor:
break
per_day = collections.Counter(v[:10] for v in first_seen.values())
print(latest, remote, f"{remote / latest:.1%}")
for day, n in sorted(per_day.items())[-14:]:
print(day, n)
A full pass means several hundred pages at 100 entries each and takes a few minutes. Be gentle: one pass a day is plenty, and the API is a shared public service.
Two details that will skew your numbers if you miss them:
isLatest, not entries.
The counter above runs as-is. Our daily snapshot job stores the result so the history cannot be lost; the snapshots behind this post start on 8 November. More small tools for agents and public data are in our data-recipes repo.
Question: does anyone know what drove the 1,311 new servers on Sunday 8 November? If it was one tool auto-publishing, I would like to understand how registries plan to deal with that.