mcpcost

report · 51 736 servers · run of 2026-10-31

The state of the MCP registry, October 2026

mcpcost probes every server in the official MCP registry once a day. It connects as a client, lists the tools, and counts what the tool definitions cost in context tokens. This report reads the daily runs of October 2026, from the first full run on 2026-10-03 to the run of 2026-10-31. Package servers from npm and PyPI joined the probe on 2026-10-15. The tables say which numbers cover them.

1. The registry in numbers

The registry lists an entry per server version. mcpcost takes the latest version of each server. It probes every entry with a remote URL, and every entry whose only package is on npm or PyPI. Entries with only an OCI or MCPB package are not probed.

Registry, 2026-10-31EntriesShare
Entries, latest versions53 549100 %
With a remote URL35 62167 %
With an npm or PyPI package only15 04728 %
Probed51 73697 %
of which remote35 54369 %
of which package, in a container16 19331 %
Hosts26 496

By result

Three groups. Alive: the server answers initialize and tools/list. Gated: the server is reachable, but it wants auth, a payment, an environment variable or an argument first. Down: everything else.

ResultGroupServersShare of probed
Alivealive30 10958 %
Gatedgated12 99925 %
Down or brokendown8 62817 %
alivealive30 10958 %
needs authgated10 18120 %
crashed on startdown3 0296 %
not probed1 8134 %
not founddown1 4643 %
needs envgated9712 %
timeoutdown8842 %
no DNSdown6721 %
install faileddown5881 %
server errordown5271 %
paid (402)gated5071 %
needs argsgated4801 %
not MCPdown4541 %
TLS errordown3131 %
needs your URLgated2721 %
rate limiteddown2671 %
needs networkgated2240 %
needs mcp<2gated2210 %
no executabledown2120 %
errordown1840 %
newer protocolgated1280 %
refuseddown340 %
rejects our protocol versiongated150 %

2. What changed in October

The first full run was on 2026-10-03. The table compares it with the run of 2026-10-31. The rows cover remote servers only, so the two runs measure the same thing. Package servers are in the second table.

Remote servers2026-10-032026-10-31Change
Registry entries, latest versions38 94253 54914 607
Entries with a remote URL24 29435 54311 249
Alive12 97520 8667 891
Alive, share of remote53 %59 %
Need auth5 83610 1814 345
Need auth, share of remote24 %29 %
Protocol 2025-11-25, share of alive with a known version60 %61 %
Hosts16 53426 4969 962

Token cost

Exact counts start on 2026-10-05, the first run with an API key. From 2026-10-15 the alive set also holds package servers. Their tool sets are smaller, so the median goes down as they join. The estimate row covers every run.

Alive servers2026-10-032026-10-052026-10-31
Alive servers in the set12 97513 33730 109
Median tokens, Claude Opus 5.5, exact2 9092 547
90th percentile, exact13 05022 570
99th percentile, exact44 09553 090
Median tokens, o200k estimate1 7271 6781 464
Median tools per server888

Run by run

33 daily runs. A point is missing where a run could not separate remote from package servers (2026-10-15 to 2026-10-20). Hover a point for its value, or open the table.

Alive remote servers

12 96816 91720 86610-0310-3110-03 · 12 97510-03 · 12 96810-04 · 13 09210-04 · 13 10610-05 · 13 32410-05 · 13 33710-06 · 13 66110-07 · 13 43610-08 · 13 67010-09 · 14 05410-10 · 14 28810-11 · 14 45710-12 · 14 75910-13 · 14 81710-14 · 15 09710-21 · 16 74010-23 · 17 00310-24 · 17 19710-25 · 17 38610-25 · 17 69210-26 · 19 23810-27 · 19 55410-28 · 19 74410-29 · 20 08510-30 · 20 59010-31 · 20 86620 866
as a table
RunValue
2026-10-3120 866
2026-10-3020 590
2026-10-2920 085
2026-10-2819 744
2026-10-2719 554
2026-10-2619 238
2026-10-2517 692
2026-10-2517 386
2026-10-2417 197
2026-10-2317 003
2026-10-2116 740
2026-10-1415 097
2026-10-1314 817
2026-10-1214 759
2026-10-1114 457
2026-10-1014 288
2026-10-0914 054
2026-10-0813 670
2026-10-0713 436
2026-10-0613 661
2026-10-0513 337
2026-10-0513 324
2026-10-0413 106
2026-10-0413 092
2026-10-0312 968
2026-10-0312 975

Alive, share of remote servers

53.056.059.010-0310-3110-03 · 53.4 %10-03 · 53.3 %10-04 · 53.5 %10-04 · 53.6 %10-05 · 54.0 %10-05 · 53.8 %10-06 · 54.5 %10-07 · 52.9 %10-08 · 53.0 %10-09 · 54.2 %10-10 · 54.5 %10-11 · 54.5 %10-12 · 54.8 %10-13 · 54.2 %10-14 · 54.5 %10-21 · 54.1 %10-23 · 53.1 %10-24 · 53.1 %10-25 · 53.0 %10-25 · 53.6 %10-26 · 58.0 %10-27 · 58.2 %10-28 · 58.0 %10-29 · 58.2 %10-30 · 58.7 %10-31 · 58.7 %58.7 %
as a table
RunValue
2026-10-3158.7 %
2026-10-3058.7 %
2026-10-2958.2 %
2026-10-2858.0 %
2026-10-2758.2 %
2026-10-2658.0 %
2026-10-2553.6 %
2026-10-2553.0 %
2026-10-2453.1 %
2026-10-2353.1 %
2026-10-2154.1 %
2026-10-1454.5 %
2026-10-1354.2 %
2026-10-1254.8 %
2026-10-1154.5 %
2026-10-1054.5 %
2026-10-0954.2 %
2026-10-0853.0 %
2026-10-0752.9 %
2026-10-0654.5 %
2026-10-0553.8 %
2026-10-0554.0 %
2026-10-0453.6 %
2026-10-0453.5 %
2026-10-0353.3 %
2026-10-0353.4 %

Median tokens of a tool set in Claude Opus 5.5, alive servers, exact

2 2712 6352 99810-0510-3110-05 · 2 90910-06 · 2 79010-07 · 2 88010-08 · 2 84310-09 · 2 99810-10 · 2 93510-11 · 2 93410-12 · 2 93310-13 · 2 97610-14 · 2 90910-15 · 2 73710-16 · 2 65510-17 · 2 56410-18 · 2 42010-19 · 2 38610-19 · 2 36710-20 · 2 35810-21 · 2 34210-23 · 2 29510-24 · 2 28610-25 · 2 27110-25 · 2 35110-26 · 2 68910-27 · 2 64610-28 · 2 63010-29 · 2 59510-30 · 2 54710-31 · 2 5472 547
as a table
RunValue
2026-10-312 547
2026-10-302 547
2026-10-292 595
2026-10-282 630
2026-10-272 646
2026-10-262 689
2026-10-252 351
2026-10-252 271
2026-10-242 286
2026-10-232 295
2026-10-212 342
2026-10-202 358
2026-10-192 367
2026-10-192 386
2026-10-182 420
2026-10-172 564
2026-10-162 655
2026-10-152 737
2026-10-142 909
2026-10-132 976
2026-10-122 933
2026-10-112 934
2026-10-102 935
2026-10-092 998
2026-10-082 843
2026-10-072 880
2026-10-062 790
2026-10-052 909

Servers that need auth, share of remote servers

24.026.029.010-0310-3110-03 · 24.0 %10-03 · 24.8 %10-04 · 25.0 %10-04 · 25.0 %10-05 · 25.2 %10-05 · 25.2 %10-06 · 25.3 %10-07 · 25.4 %10-08 · 25.5 %10-09 · 25.9 %10-10 · 26.1 %10-11 · 26.2 %10-12 · 26.3 %10-13 · 26.4 %10-14 · 26.6 %10-15 · 26.8 %10-16 · 27.0 %10-17 · 27.2 %10-18 · 27.2 %10-19 · 27.2 %10-19 · 27.3 %10-20 · 27.4 %10-21 · 27.7 %10-23 · 27.8 %10-24 · 27.9 %10-25 · 28.1 %10-25 · 28.1 %10-26 · 28.0 %10-27 · 28.1 %10-28 · 28.5 %10-29 · 28.7 %10-30 · 28.6 %10-31 · 28.6 %28.6 %
as a table
RunValue
2026-10-3128.6 %
2026-10-3028.6 %
2026-10-2928.7 %
2026-10-2828.5 %
2026-10-2728.1 %
2026-10-2628.0 %
2026-10-2528.1 %
2026-10-2528.1 %
2026-10-2427.9 %
2026-10-2327.8 %
2026-10-2127.7 %
2026-10-2027.4 %
2026-10-1927.3 %
2026-10-1927.2 %
2026-10-1827.2 %
2026-10-1727.2 %
2026-10-1627.0 %
2026-10-1526.8 %
2026-10-1426.6 %
2026-10-1326.4 %
2026-10-1226.3 %
2026-10-1126.2 %
2026-10-1026.1 %
2026-10-0925.9 %
2026-10-0825.5 %
2026-10-0725.4 %
2026-10-0625.3 %
2026-10-0525.2 %
2026-10-0525.2 %
2026-10-0425.0 %
2026-10-0425.0 %
2026-10-0324.8 %
2026-10-0324.0 %

Protocol 2025-11-25, share of alive servers with a known version

59.062.065.010-0310-3110-03 · 59.6 %10-03 · 59.4 %10-04 · 59.7 %10-04 · 59.6 %10-05 · 60.2 %10-05 · 60.1 %10-06 · 60.9 %10-07 · 59.8 %10-08 · 60.5 %10-09 · 59.7 %10-10 · 60.4 %10-11 · 60.6 %10-12 · 60.6 %10-13 · 60.3 %10-14 · 60.7 %10-15 · 61.5 %10-16 · 61.3 %10-17 · 61.6 %10-18 · 62.0 %10-19 · 62.3 %10-19 · 62.5 %10-20 · 62.5 %10-21 · 62.8 %10-23 · 64.5 %10-24 · 64.7 %10-25 · 64.8 %10-25 · 65.3 %10-26 · 60.0 %10-27 · 60.1 %10-28 · 60.2 %10-29 · 60.5 %10-30 · 60.5 %10-31 · 60.6 %60.6 %
as a table
RunValue
2026-10-3160.6 %
2026-10-3060.5 %
2026-10-2960.5 %
2026-10-2860.2 %
2026-10-2760.1 %
2026-10-2660.0 %
2026-10-2565.3 %
2026-10-2564.8 %
2026-10-2464.7 %
2026-10-2364.5 %
2026-10-2162.8 %
2026-10-2062.5 %
2026-10-1962.5 %
2026-10-1962.3 %
2026-10-1862.0 %
2026-10-1761.6 %
2026-10-1661.3 %
2026-10-1561.5 %
2026-10-1460.7 %
2026-10-1360.3 %
2026-10-1260.6 %
2026-10-1160.6 %
2026-10-1060.4 %
2026-10-0959.7 %
2026-10-0860.5 %
2026-10-0759.8 %
2026-10-0660.9 %
2026-10-0560.1 %
2026-10-0560.2 %
2026-10-0459.6 %
2026-10-0459.7 %
2026-10-0359.4 %
2026-10-0359.6 %

Registry entries, latest versions

38 94246 24653 54910-0310-3110-03 · 38 94210-03 · 38 98810-04 · 39 13810-04 · 39 15910-05 · 39 40710-05 · 39 61410-06 · 39 94310-07 · 40 32810-08 · 40 82210-09 · 41 05110-10 · 41 40710-11 · 41 78410-12 · 38 94210-13 · 42 79710-14 · 43 29110-15 · 43 84510-16 · 44 46210-17 · 45 12310-18 · 45 67810-19 · 46 18410-19 · 46 39310-20 · 46 82410-21 · 47 50710-23 · 48 74010-24 · 49 22110-25 · 49 73410-25 · 49 99210-26 · 50 29710-27 · 50 89710-28 · 51 69610-29 · 52 30310-30 · 52 98810-31 · 53 54953 549
as a table
RunValue
2026-10-3153 549
2026-10-3052 988
2026-10-2952 303
2026-10-2851 696
2026-10-2750 897
2026-10-2650 297
2026-10-2549 992
2026-10-2549 734
2026-10-2449 221
2026-10-2348 740
2026-10-2147 507
2026-10-2046 824
2026-10-1946 393
2026-10-1946 184
2026-10-1845 678
2026-10-1745 123
2026-10-1644 462
2026-10-1543 845
2026-10-1443 291
2026-10-1342 797
2026-10-1238 942
2026-10-1141 784
2026-10-1041 407
2026-10-0941 051
2026-10-0840 822
2026-10-0740 328
2026-10-0639 943
2026-10-0539 614
2026-10-0539 407
2026-10-0439 159
2026-10-0439 138
2026-10-0338 988
2026-10-0338 942

3. Token cost

A client sends the model the name, the description and the input schema of each tool. It does this in each request, also when the model calls no tool. The numbers below are the cost of the tool definitions of one server, as the Anthropic count_tokens endpoint counts them for Claude Opus 5.5. The endpoint is free and calls no model. Claude Code puts mcp__<server>__ in front of each tool name. The second column counts that variant.

Alive servers, 28 481 countedClaude Opus 5.5In Claude Codeo200k estimate
Median2 5472 6371 464
90th percentile22 57023 11012 840
99th percentile53 09054 23531 810
Maximum1 633 6241 634 208851 305
Sum over all counted servers252 386 178257 765 772155 921 548

The estimate column covers all 30 109 alive servers. 366 alive servers publish a schema the endpoint refuses; they have the estimate only.

Servers per cost band

How many of the counted servers fall into each band of exact tokens.

  1. 2 462< 500
  2. 3 799500 – 1k
  3. 5 3361k – 2k
  4. 6 8292k – 5k
  5. 2 9575k – 10k
  6. 4 05210k – 20k
  7. 2 73920k – 50k
  8. 20050k – 100k
  9. 107100k+
as a table
TokensServers
< 5002 462
500 – 1k3 799
1k – 2k5 336
2k – 5k6 829
5k – 10k2 957
10k – 20k4 052
20k – 50k2 739
50k – 100k200
100k+107

Exact against the estimate

The estimate serializes each tool as compact JSON and counts it with the o200k_base tokenizer. Claude renders the schemas in its own format, so the exact count is higher. The ratio is steady across servers.

Exact count divided by the estimateRatio
10th percentile1.63
Median1.75
90th percentile1.87

The heaviest tool sets

Claude's context window holds 200 000 tokens. The last column says how much of it one server takes before the first prompt. At least 20 alive servers do not fit into the window at all.

ServerHostToolsClaude Opus 5.5In Claude CodeOf 200k
com.replynodes/mcpmcp.replynodes.com731 633 6241 634 208817 %
com.mcparmory/posthogpypi.org6181 025 2561 030 200513 %
info.astroway/astrologymcp.astroway.info799927 903934 295464 %
io.github.WellApp-ai/well-mcpapi.wellapp.ai424796 849801 089398 %
co.ainumbers/toolsmcp.ainumbers.co754783 386788 664392 %
io.github.devinwang/app-store-connect-mcpnpmjs.com1 316660 300680 040330 %
com.idosgames/platformsite.idosgames.com159610 144611 257305 %
net.crawlora/crawlora-mcpmcp.crawlora.net2 000596 269622 269298 %
io.github.whiteknightonhorse/apibaseapibase.pro1 402592 891604 107296 %
com.mcparmory/elevenlabspypi.org258529 755532 593265 %
com.mcparmory/atlassian-jirapypi.org317438 661442 465219 %
com.mcparmory/datadogpypi.org770425 131431 291213 %

All servers by token cost → · What a server costs, the short answer →

4. Auth

A server that answers 401 or 403 is reachable, but it does not talk to a client without credentials. The probe records the WWW-Authenticate scheme. Then it reads what the server publishes about getting in: the OAuth protected-resource metadata (RFC 9728) and the authorization server's metadata (RFC 8414). It never asks for a token.

WWW-Authenticate schemeServersShare of servers that need auth
Bearer8 94988 %
no WWW-Authenticate header1 19512 %
ApiKey200 %
AzureApiManagementKey60 %
Api-Key20 %
OAuth20 %
other schemes, 6 of them7

How a client gets in

Four kinds. The sentence under each one says what it means for a client.

KindServersShareFor a client
oauth-dcrOAuth, clients register themselves8 21981 %The authorization server registers clients itself (RFC 7591). Claude Code, Cursor or npx mcpcost auth connect with no setup. The user signs in and consents.
oauthOAuth, the publisher issues the client id5295 %The server publishes OAuth metadata, but the publisher issues the client id. A client needs a registration with the publisher before it can start the flow.
tokena key or token from the publisher4695 %The server wants a key or a token in a header. The publisher gives it out. A client has no standard way to get one.
unknownno WWW-Authenticate header, no OAuth metadata9649 %The server answers 401 or 403 and publishes nothing a client can read: no WWW-Authenticate scheme, no OAuth metadata. A client can only read the documentation.

All servers that need auth → · How the kinds are read →

5. The lumps

The registry is not flat. Most hosts publish one server. A few hosts publish thousands of entries from one template. The table lists the hosts with at least 200 entries. “Distinct tool sets” is the number of different tool lists one host serves: a host with many entries and few tool sets publishes the same server under many names.

HostEntriesAliveGatedDownDistinct tool setsMedian tokens, estimate
agent-observatory-sensor.nolimit-observatory.workers.dev2 3652 365002 3417 424
gateway.pipeworx.io1 9091 8370551 83526 653
api.mcp.ai1 1171 117001 0441 241
github.com90931337553051 751
mcp.apify.com6958684081 835
ghcr.io4403104292931 536
api.m2mcent.com276027600
server.smithery.ai2160261900

Duplicate tool sets

A hash of the sorted tool names and schema sizes marks servers that serve the same tools under different names.

Alive serversServersShare
Alive servers30 109100 %
Distinct tool sets28 88896 %
Servers that copy another server's tool set1 2214 %

Packages

The two package registries are the biggest hosts. A package is probed in a container with the network off and placeholder values for the environment variables the registry entry lists.

HostEntriesAliveGatedDownDistinct tool setsMedian tokens, estimate
npmjs.com10 3887 4379451 8027 169955
pypi.org4 4011 8068651 6282 0031 847

The 500 biggest hosts hold 25 562 of the 51 736 probed servers. The other 25 996 hosts hold the rest. All hosts →

6. Protocol versions

The version a server negotiates in initialize. The container run does not record the version of a package server, so package servers are the “unknown” row. A server that answers with a version the current SDK rejects is counted as “newer protocol” in section 1, not here.

Negotiated versionAlive serversShare of alive with a known version
2025-11-2512 23061 %
2025-03-264 82724 %
2025-06-182 51112 %
2024-11-056013 %
unknown9 940

128 servers answer with a version the SDK does not support yet. They are reachable and count as gated.

7. What the registry does not catch

The registry checks who may publish under a name. It does not check that the endpoint speaks MCP, or that it still exists. These servers are listed as active, and a client that adds one gets an error. Not MCP, not found, no DNS and paid are the big ones.

ResultWhat the probe sawServersShare of remote
not foundHTTP 404, 405 or 410 on the endpoint the registry names1 4644 %
timeoutno answer in 15 seconds8842 %
no DNSthe host name does not resolve6722 %
server errorHTTP 5xx5271 %
paid (402)HTTP 402: the server wants a payment before it talks5071 %
not MCPHTTP 200, but the body is not a valid MCP reply, or a JSON-RPC error to initialize4541 %
TLS errorthe certificate or the handshake fails3131 %
rate limitedHTTP 429 to the first request2671 %
erroranother transport failure1841 %
refusedthe connection is refused340 %
Together5 30615 %

Packages that do not start

ResultWhat the probe sawPackagesShare of packages
crashed on startthe package starts and exits before it answers initialize3 02919 %
install failednpm or pip cannot install the package5884 %
no executablethe package installs but has no executable to start2121 %

775 PyPI package servers stop at the import of mcp.server.fastmcp. mcp 2.0, the Python SDK of 28 July 2026, renamed FastMCP to MCPServer, and these packages ask for mcp with no upper bound, so a plain install and uvx get 2.x. Since 31 October the probe installs such a package a second time with mcp<2, and 221 of them started that way by this run and listed their tools: needs mcp<2. The probe checks each package once a week, so this number grows until each of them had its second install. The fix for a publisher is one line: pin mcp<2 (for example mcp>=1.2,<2), or move to MCPServer.

What changed since yesterday → · All servers that are down →

8. Method and limits

The probe connects to each server once a day with the official TypeScript SDK, sends initialize and tools/list, and sends no credentials. The exact token count comes from the Anthropic count_tokens endpoint: one request with the server's tools and a marker tool, one with the marker tool only, and the difference is the cost. A server is counted again only when its tool definitions change. A server that needs auth shows no tools, so its cost is not in the numbers. One probe from one network on one day can miss a server that is alive: a timeout or a rate limit on that day counts as down. Package servers run in a container with the network off, so a package that needs a real network or a real credential shows as gated, not as alive. A PyPI package that stops at the import of FastMCP from mcp 2.x is installed a second time with mcp<2, and when it then lists its tools it is gated (needs mcp<2), with its tools and token counts.

The full method → · The source →

9. The dataset

One row per probed server from the run of 2026-10-31, with the numbers the board shows for it. The numbers of a row are the numbers the JSON API gives for that server, /api/s/<name>.json, on that day. The file is a static file of this site: it is small, and a static file costs no Worker request on the free tier, so it is the cheaper of the two ways to publish it. A GitHub release would hold a file over 10 MB.

51 736rows, one per probed server
13fields per row
1.6 MBgzip of a JSON document
CC BY 4.0use it, name the source

Download mcpcost-2026-10.json.gz

The document has the licence, the run date, the field list and the citation at the top. The rows are under rows. A missing value is null.

curl -sL https://mcpcost.ulehla.dev/data/mcpcost-2026-10.json.gz | gunzip | jq '.rows[] | select(.status == "ok") | .claude_tokens' | sort -n | tail -3

Fields

FieldMeaning
namethe registry name; the server page is /s/<name>
hostthe host of the remote endpoint; npmjs.com or pypi.org for a package server
transportstreamable-http, sse or stdio
statusthe probe result, the same words as the status filter of the board
http_statusthe HTTP status of the first response, when there was one
auth_schemethe WWW-Authenticate scheme of a server that needs auth
auth_kindhow a client gets in: oauth-dcr, oauth, token or unknown
protocol_versionthe protocol version the server negotiated
tool_countthe number of tools in tools/list
tools_tokensestimate: o200k_base over the compact JSON of the tool definitions
claude_tokensexact: the Anthropic count_tokens endpoint, claude-opus-5-5, tool names as published
claude_tokens_ccexact: the same with the Claude Code prefix mcp__<server>__ on each tool name
probed_atwhen the run probed the server; a package server keeps the time of its last container run

The same run as re-check evidence for the registry

The official registry discusses a periodic re-check of its entries (issue #1715). This file is this run in the row shape agreed there, so a registry job can merge it with evidence from other sources: entry, checked_at, check (endpoint for a remote server, package for an npm or PyPI server started in a container), state and evidence. The state is one of five: alive (initialize and tools/list succeeded), auth (reachable, needs credentials or payment), transient (no valid answer in this run: a timeout, 429, 5xx or a refused connection, or a dead cause seen for less than two days), dead (no DNS, TLS failure, 404, not MCP, in every check for two days or more; a package that does not install), unknown (a package that needs values or a network to start, or crashed on start; a package that starts only with an older major version of its SDK, mcp<2, status needs_old_sdk; or a URL with placeholders that each user fills in). Each row also has the entry version when the entry did not change after the check. The evidence holds the probe status by cause, the target, the HTTP status, the auth scheme and kind, the protocol version, the tool count, the first failed check and the first line of the error.

Corrected on 2026-10-26. The first file called 191 entries dead whose URL has placeholders that each user fills in, such as https://{tenant}.example.com/mcp, and it called a dead cause dead after one failed check. Now such a URL is unknown, and a dead cause counts after two days of failed checks, from the runs since 2026-10-18: 1 206 rows fewer are dead. The board follows the same rules from 2026-10-26.

Download mcpcost-evidence-2026-10.json.gz (2.1 MB, CC BY 4.0, the same licence and citation)

The newest run in the same shape is always at /data/mcpcost-evidence-latest.json.gz. It changes after each daily run. Its index, mcpcost-evidence-latest.index.json, gives the run date, the number of rows, the size and the SHA-256.

Licence and citation

The dataset is published under Creative Commons Attribution 4.0. You may copy it, change it and build on it, also for commercial work. Name the source. A link to this page is enough.

Ondřej Úlehla, The state of the MCP registry, October 2026. mcpcost, run of 2026-10-31. https://mcpcost.ulehla.dev/report/2026-10. CC BY 4.0.

The code of the probe and of this site is MIT. The daily summaries since 2026-10-03 are committed in the repository as data/latest/summary.json, and the summary API serves the latest one.