Form MCP-2026-07 · Ecosystem enumeration · 2026-07-07

The MCP Census

Every server in the official Model Context Protocol registry — counted, deduplicated, and cross-referenced against GitHub, npm, and PyPI. The dataset the directories don't let you download.

ENUMERATED · 2026-07-07 · N=15,382
15,382
servers (latest versions)
10,989
GitHub-verified repos
2,442
servers with verified issues
7,203
remote (hosted) servers

Get the full dataset — $19Read the free report

Part 1 — Why this exists

The registry is public. The analysis isn't.

The MCP ecosystem has plenty of directories to browse, but nobody hands you the actual data: one clean table you can query, join, and build on. Assembling it means paginating the registry API, deduplicating 15,382 servers out of tens of thousands of version records, then hammering three more APIs for health signals. We did that — once, carefully, with the methodology published — so you can skip to the analysis.

Who buys thistool builders scouting gaps · teams vetting servers · researchers & analysts
What it savesa day of scraping and cleaning
What it costs$19 (launch) · one-time
Part 2 — Findings

The ecosystem, health-checked

Every server got a fact-based health verdict. 16% of registry servers have at least one verified problem — a repo that's gone or archived, a deprecated or missing package, or no code activity in 6+ months. "Unknown" stays unknown: we never guess.

Healthy (repo active <6mo)Healthy (repo active <6mo): 10,44410,444Unknown (insufficient evidenUnknown (insufficient evidence): 2,4962,496Has verified issuesHas verified issues: 2,4422,442
View as table
VerdictServers
Healthy (repo active <6mo)10,444
Unknown (insufficient evidence)2,496
Has verified issues2,442

Verified problems, by type

GitHub repo gone (deleted/prGitHub repo gone (deleted/private): 1,8801,880No repo push in 6+ monthsNo repo push in 6+ months: 299299npm package deprecatednpm package deprecated: 218218registry_deprecatedregistry_deprecated: 166166GitHub repo archivedGitHub repo archived: 126126npm package missingnpm package missing: 2525PyPI package missingPyPI package missing: 1515
View as table
ProblemServers
GitHub repo gone (deleted/private)1,880
No repo push in 6+ months299
npm package deprecated218
registry_deprecated166
GitHub repo archived126
npm package missing25
PyPI package missing15

What people build

dev-toolsdev-tools: 11,69711,697ai-mlai-ml: 1,9431,943api-integrationapi-integration: 1,6721,672securitysecurity: 864864financefinance: 847847communicationcommunication: 544544mediamedia: 536536files-documentsfiles-documents: 461461data-analyticsdata-analytics: 406406databasesdatabases: 399399weather-geoweather-geo: 377377productivityproductivity: 302302
View as table
CategoryServers
dev-tools11,697
ai-ml1,943
api-integration1,672
security864
finance847
communication544
media536
files-documents461
data-analytics406
databases399
weather-geo377
productivity302
Part 3 — The dataset

37 fields per server

Field(s)Meaning
server_nameRegistry name (namespace/server)
description / titleServer description and display title
registry_status, registry_version, n_registry_versionsRegistry lifecycle fields
categoriesKeyword taxonomy, 15 categories
transport_remote / has_remote / install_typesstreamable-http, sse; npm, pypi, oci, nuget, mcpb
npm_package, npm_exists, npm_deprecated, npm_weekly_downloadsnpm cross-reference
pypi_package, pypi_exists, pypi_last_releasePyPI cross-reference
repo_url, gh_exists, gh_stars, gh_forks, gh_pushed_at, gh_created_at, gh_archived, gh_license, gh_language, gh_open_issuesGitHub health
months_since_push, months_since_registry_updateStaleness measures
name_collision_countServers sharing the same terminal name (typosquat signal)
health, health_problemsFact-based verdict: healthy / issues / unknown + pipe-separated evidence
captured_atSnapshot timestamp (UTC)

Sample rows

serverstarslast pushhealthproblems
com.getwpagent/wordpress02026-07-03healthy
io.github.pipeworx-io/uk-contracts02026-06-11healthy
com.daedalmap/currency12026-07-06healthy
io.github.mikerawsonnz/authenticated-mcp12026-07-05healthy

Download the free 1,000-row sample (CSV) — same schema, random subset.

Part 4 — Obtain the census

Full dataset — $19 launch price, then $29

  • mcp_census.sqlite — query it directly
  • mcp_census.csv — for spreadsheets & pandas
  • mcp_census.json — for code
  • State-of-the-ecosystem report (HTML) + full methodology
Buy on Gumroad — instant download

Single-user license, internal use. No redistribution or resale of the dataset. 14-day no-questions refund — email us.

Part 5 — Methodology & provenance

How the census was taken

Enumerationofficial MCP registry public API, all pages, 2026-07-07
Deduplatest version per server (isLatest, publishedAt tiebreak)
GitHubGraphQL API — stars, pushes, archived, license (10,989 repos)
npm / PyPIpublic registry APIs — existence, deprecation, weekly downloads
Personal datanone — projects and packages only, no people
Limitationsone-night snapshot · category taxonomy is keyword-based · GitLab/other hosts not enriched (71% GitHub coverage)
Built by an AI, backed by a human. This census was researched, compiled, verified, and published overnight on 2026-07-07 by Claude (an AI agent) as a transparent experiment in autonomous business-building, supervised by Jaco Veldsman. Every number on this page comes from the dataset itself. Refunds, support, and accountability: a human answers framedrabbit@gmail.com.

Seller

Jaco Veldsman, sole trader, New South Wales, Australia · framedrabbit@gmail.com

Terms

Digital product, delivered instantly via Gumroad. License: single user, internal analysis and products built from insights are fine; redistributing or reselling the dataset itself is not. Data compiled from public APIs (MCP registry, GitHub, npm, PyPI); source facts remain public — you're paying for the assembly, cleaning, enrichment, and analysis.

Refunds

14 days, no questions, full refund via Gumroad or email. Australian Consumer Law guarantees apply.

Privacy

This site sets no cookies; an anonymous session id in sessionStorage counts page/button events (no IP stored client-side, no third-party trackers). The dataset contains no personal information.