MCP server statistics
How many Model Context Protocol servers were first published each month, how many entries have been updated since, and which source each of the 29,756 servers in this catalogue came from. Counted from the records on 8 September 2026, with the caveats that decide how much any of it means.
How the catalogue has grown
Both charts below count every server in the catalogue, whichever source it came from. They are drawn from two different dates on the same record and answer two different questions, so they are worth reading as a pair rather than as one number twice.
Servers first published per month
The month the source records each server as first published, not the month we crawled it. A publisher can register something they wrote a year earlier, so this counts registry entries created rather than software written.
Two numbers on one chart. The bars are how many arrived that month, read against the left axis. The line is how many the catalogue held in total by the end of it, read against the right.
The dotted bars and dotted line are an estimate, not a measurement: 3,880 a month for 6 months, reaching 53,036 by March 2027. A flat mean of the 6 complete months from 2026-03 to 2026-08. The current month is left out because it is partial, and no curve is fitted because the catalogue has grown in steps as whole sources were added, which a fitted trend would turn into a confident wrong number.
| Month | First published | Running total | Measured or estimated |
|---|---|---|---|
| February 2025 | 1 | 1 | measured |
| March 2025 | 1 | 2 | measured |
| April 2025 | 1 | 3 | measured |
| May 2025 | 44 | 47 | measured |
| June 2025 | 5 | 52 | measured |
| July 2025 | 21 | 73 | measured |
| August 2025 | 23 | 96 | measured |
| September 2025 | 402 | 498 | measured |
| October 2025 | 276 | 774 | measured |
| November 2025 | 198 | 972 | measured |
| December 2025 | 243 | 1,215 | measured |
| January 2026 | 316 | 1,531 | measured |
| February 2026 | 975 | 2,506 | measured |
| March 2026 | 1,758 | 4,264 | measured |
| April 2026 | 2,126 | 6,390 | measured |
| May 2026 | 2,590 | 8,980 | measured |
| June 2026 | 3,527 | 12,507 | measured |
| July 2026 | 4,833 | 17,340 | measured |
| August 2026 | 8,444 | 25,784 | measured |
| September 2026 | 3,972 | 29,756 | measured |
| October 2026 | 3,880 | 33,636 | estimate |
| November 2026 | 3,880 | 37,516 | estimate |
| December 2026 | 3,880 | 41,396 | estimate |
| January 2027 | 3,880 | 45,276 | estimate |
| February 2027 | 3,880 | 49,156 | estimate |
| March 2027 | 3,880 | 53,036 | estimate |
The last month in the chart is still in progress on 8 September 2026, so its bar is short for a reason that has nothing to do with the ecosystem. All 29,756 records carry a first published date, so nothing is missing from the chart.
Servers by the month their entry was last updated
The month each server's registry entry was last touched, again one server per row. A server appears in exactly one month here, its most recent one, so this shows how much of the catalogue is still being maintained and how much has gone quiet. It is deliberately labelled as last updated rather than as updates, because it cannot count how many times a server changed.
| Month | Last updated in that month | Share of catalogue |
|---|---|---|
| February 2025 | 1 | 0.0% |
| March 2025 | 1 | 0.0% |
| April 2025 | 1 | 0.0% |
| May 2025 | 44 | 0.1% |
| June 2025 | 5 | 0.0% |
| July 2025 | 21 | 0.1% |
| August 2025 | 23 | 0.1% |
| September 2025 | 401 | 1.3% |
| October 2025 | 275 | 0.9% |
| November 2025 | 198 | 0.7% |
| December 2025 | 243 | 0.8% |
| January 2026 | 315 | 1.1% |
| February 2026 | 898 | 3.0% |
| March 2026 | 1,718 | 5.8% |
| April 2026 | 2,121 | 7.1% |
| May 2026 | 2,579 | 8.7% |
| June 2026 | 3,613 | 12.1% |
| July 2026 | 4,836 | 16.3% |
| August 2026 | 8,476 | 28.5% |
| September 2026 | 3,987 | 13.4% |
Read the two charts together and the difference is the interesting part. A month whose last updated bar is much shorter than its first published bar is a month whose servers have mostly been revised since. A month where the two are nearly equal is a month whose entries were registered and then left alone.
The same two series, per source
The official registry is the spine of this catalogue and everything else is supplementary. Each record is counted once, under the source that established its identity, so the four figures below add up to the catalogue rather than overlapping. The three charts that follow keep their own scale: the registry is two orders of magnitude larger than Docker's catalogue, and a shared axis would flatten the smaller two into nothing.
| Source | Servers | Share of catalogue | First published since Jul 2026 |
|---|---|---|---|
| Official MCP registry | 28,298 | 95.1% | 16,850 |
| Smithery | 1,297 | 4.4% | 394 |
| Docker MCP catalogue | 156 | 0.5% | 0 |
| Hand added | 5 | 0.0% | 5 |
Official MCP registry, 28,298 servers
Servers whose identity comes from the official MCP registry. This is the only source with a publish date and an update date for every entry, so it is the only one of the three whose curve means much on its own.
- First published
- Last updated
| Month | First published | Last updated |
|---|---|---|
| February 2025 | 0 | 0 |
| March 2025 | 0 | 0 |
| April 2025 | 0 | 0 |
| May 2025 | 0 | 0 |
| June 2025 | 0 | 0 |
| July 2025 | 0 | 0 |
| August 2025 | 0 | 0 |
| September 2025 | 344 | 343 |
| October 2025 | 228 | 227 |
| November 2025 | 139 | 139 |
| December 2025 | 225 | 225 |
| January 2026 | 289 | 288 |
| February 2026 | 906 | 829 |
| March 2026 | 1,585 | 1,545 |
| April 2026 | 1,936 | 1,931 |
| May 2026 | 2,422 | 2,411 |
| June 2026 | 3,374 | 3,460 |
| July 2026 | 4,737 | 4,740 |
| August 2026 | 8,174 | 8,206 |
| September 2026 | 3,939 | 3,954 |
Smithery, 1,297 servers
Servers this catalogue holds because Smithery lists them and the registry does not. Smithery publishes tool schemas at scale, which is why it is worth joining, but its pagination caps at 500 records per query, so this is a sample of its index rather than all of it. The shape of this series says as much about our crawl as about Smithery.
- First published
- Last updated
| Month | First published | Last updated |
|---|---|---|
| February 2025 | 1 | 1 |
| March 2025 | 1 | 1 |
| April 2025 | 1 | 1 |
| May 2025 | 3 | 3 |
| June 2025 | 1 | 1 |
| July 2025 | 6 | 6 |
| August 2025 | 13 | 13 |
| September 2025 | 16 | 16 |
| October 2025 | 45 | 45 |
| November 2025 | 25 | 25 |
| December 2025 | 17 | 17 |
| January 2026 | 26 | 26 |
| February 2026 | 67 | 67 |
| March 2026 | 173 | 173 |
| April 2026 | 188 | 188 |
| May 2026 | 168 | 168 |
| June 2026 | 152 | 152 |
| July 2026 | 95 | 95 |
| August 2026 | 266 | 266 |
| September 2026 | 33 | 33 |
Docker MCP catalogue, 156 servers
Servers this catalogue holds because Docker's MCP catalogue lists them and the registry does not. That catalogue is curated rather than open to submissions, and small, so single months of 1 or 2 are normal here and a flat month is not a stall.
- First published
- Last updated
| Month | First published | Last updated |
|---|---|---|
| February 2025 | 0 | 0 |
| March 2025 | 0 | 0 |
| April 2025 | 0 | 0 |
| May 2025 | 41 | 41 |
| June 2025 | 4 | 4 |
| July 2025 | 15 | 15 |
| August 2025 | 10 | 10 |
| September 2025 | 42 | 42 |
| October 2025 | 3 | 3 |
| November 2025 | 34 | 34 |
| December 2025 | 1 | 1 |
| January 2026 | 1 | 1 |
| February 2026 | 2 | 2 |
| March 2026 | 0 | 0 |
| April 2026 | 2 | 2 |
| May 2026 | 0 | 0 |
| June 2026 | 1 | 1 |
| July 2026 | 0 | 0 |
| August 2026 | 0 | 0 |
| September 2026 | 0 | 0 |
When we first listed each server
Every record also carries the date this catalogue first held it, which is a different fact again from either date above. It is published here because it is the honest series for the question "what did this site add", and because it must not be mistaken for the registry's own history. It measures our collection schedule: the first month holds the entire catalogue as it stood when the field was introduced, and later months hold whatever a run picked up.
| Month | First listed | Share of catalogue |
|---|---|---|
| August 2026 | 20,838 | 70.0% |
| September 2026 | 8,918 | 30.0% |
So a reader wanting the ecosystem should use the first published series, and a reader wanting to know what is new on this site should use this one. They will not agree, and neither is a correction of the other.
How to read these numbers
A spike can be one publisher rather than growth
The registry does not limit how many entries one publisher may create, and several have registered large batches in a single day. That lands on this page as a tall bar, which reads as the ecosystem accelerating when it is one namespace filling in. The table below names the largest publisher namespace in every month where one namespace accounted for at least 10 percent of the registry entries first published that month, so a bar can be checked instead of trusted. Servers that reached this catalogue through Smithery or Docker are left out of these figures, because they carry no registry namespace to group by.
| Month | Registry entries | Largest namespace | Its entries | Its share |
|---|---|---|---|---|
| September 2025 | 344 | ai.smithery | 124 | 36.0% |
| October 2025 | 228 | ai.smithery | 87 | 38.2% |
| April 2026 | 1,936 | io.github.pipeworx-io | 238 | 12.3% |
| May 2026 | 2,422 | io.github.pipeworx-io | 278 | 11.5% |
| June 2026 | 3,374 | io.github.pipeworx-io | 579 | 17.2% |
| August 2026 | 8,174 | io.github.mcp-dir | 1,004 | 12.3% |
The largest single case is August 2026, where 1,004 of the 8,174 registry entries first published that month share the namespace io.github.mcp-dir. Naming it states what the registry record says and nothing further: a publisher is entitled to register as many entries as it wants to, and we make no claim about what any of them do.
Published is not written, and it is not released
The first published date is the moment the registry accepted an entry. Plenty of MCP servers existed for months before their publisher got round to registering one, and a few registered an entry before there was much to install. Treat the series as the registry's intake, which is what it is, rather than as the date the software appeared.
There is no honest count of versions published per month
The registry's primary key is a name and a version, so counting versions per month needs one row per version. The dump we hold does carry per version dates, and that is exactly why we checked it rather than assumed: it holds 9,095 version rows across 3,503 servers, of which 1,228 have more than one version, and its newest version date is 3 August 2026. That is a fraction of the 29,756 servers here, and the crawl behind it stopped part way through the alphabet, between ac.inference.sh/inference and com.stagenth/memory-kit. A monthly series drawn from it would collapse after 3 August 2026 for reasons that have nothing to do with the registry, so we do not draw one, and the second chart on this page is labelled last updated rather than updates.
What that dump does support is published here rather than thrown away, with its limits in the caption. Read it as the shape of version activity among 3,503 servers whose names sort before com.stagenth/memory-kit, and not as a catalogue wide figure.
| Month | Versions published |
|---|---|
| September 2025 | 304 |
| October 2025 | 200 |
| November 2025 | 188 |
| December 2025 | 59 |
| January 2026 | 102 |
| February 2026 | 327 |
| March 2026 | 527 |
| April 2026 | 865 |
| May 2026 | 1,108 |
| June 2026 | 1,901 |
| July 2026 | 3,237 |
| August 2026 | 277 |
The newest month is always incomplete
This page was generated on 8 September 2026, so September 2026 holds only the part of the month that had happened by then. Nothing is projected forward to fill it. The rolling 3 month figure quoted at the top and on the home page has the same edge, and says so in both places.
Checking any of this
Every figure here is counted from the JSON records in this catalogue and written to one file before the page is built, so the page states numbers rather than working them out. If one of them is wrong, the aggregate is wrong and it is rebuilt by a single command. Corrections are welcome, and a publisher correcting their own entry in the registry fixes it here on the next pass.
Frequently asked questions
- How many MCP servers are there?
- This catalogue holds 29,756 servers as of 8 September 2026. That is what the official MCP registry, Docker's MCP catalogue and Smithery between them list publicly, not a census of every server that exists: a server that is never listed anywhere cannot be counted.
- How fast is the MCP ecosystem growing?
- 17,249 servers carry a first published date in the 3 months from July 2026 to September 2026, which is 58.0% of the catalogue. Read that as the rate at which entries are being registered rather than the rate at which servers are being written, and note that the newest month is still in progress.
- Does this page count updates per month?
- No. Each server appears once, in the month its registry entry was last touched, so the second chart counts servers whose most recent change falls in that month. A true count of versions published per month needs one row per version, and the version history we hold covers 3,503 servers rather than the whole catalogue, so we do not draw it.
- Why would one month be much higher than the month before it?
- Usually because one publisher registered a large number of entries at once. This page has a table naming the largest publisher namespace in every month where one namespace accounted for at least 10 percent of that month's registry entries, so a jump can be checked rather than assumed to be growth.
- Where do these numbers come from?
- They are counted from the 29,756 JSON records in this catalogue by scripts/cli.py stats, and written to one file that both this page and the home page read. No figure on this page is calculated while the page is rendered, so the page and the data cannot disagree.
- Are the earliest months complete?
- The first published series starts in February 2025 and the counts before September 2025 are very small, because the official registry itself was new. Those months describe the registry's own history, not the history of the protocol.