Public-signal adoption data for MCP servers

See which MCP servers the ecosystem actually runs.

Search every server and client by real adoption — measured from the public repositories that wire them up. The reliability layer on top of the catalog, not another directory.

13,665
servers catalogued
6,605
observed in real repos
5,890
public repositories
18,218
usage edges mapped

Search 13,665 servers and every client observed in the wild — or scan your own stack.

For server authors & vendors

See who's adopting you — and your competitors.

Every server has a live page measured from the public repos that actually wire it up. Track your adoption, see which clients install you, claim the page to add your author band, and drop a badge in your README that stays current.

  • See exactly which public repos and orgs wire you up — and export the list
  • Claim your page: safer-mode flags, intended scopes, recommended config
  • An embeddable badge that updates as adoption changes
  • Reliability & permission signal from real installs (opt-in ax-ray)

Live adoption badge — paste into any README

adoption badge[![ax-ray](/badge/npm/%40upstash/context7-mcp.svg?metric=adoption)](/server/npm/%40upstash/context7-mcp)

Live for @upstash/context7-mcp — counts public repos observed wiring it up.

For teams adopting MCP

Choose a server on evidence, not vibes.

Rank candidates by adoption people can verify, see what teams who run one server also wire up, and scan your own repo to inventory every MCP server it pulls in — before you ship it.

  • Rank by repos observed, weekly downloads, and stars
  • “Teams using X also use Y” co-occurrence signal
  • Scan a repo or config for its full MCP footprint
  • Save scans and revisit them as the signals move
@upstash/context7-mcp
● center: this server━ inner ring: co-occurring servers (size = shared repos)┄ outer ring: consumer repos— hover a node for its name, click to open
For platform & security teams

Audit your MCP supply chain.

Point the scanner at any public repo and get a report on every server it wires up — resolved to one identity across all six kinds — with flags for unclaimed, stale, or missing-SECURITY.md. And see community-observed reliability and permission profiles, aggregated from opt-in ax-ray runs. Observations, never verdicts; every signal is re-derivable or k-anonymized.

  • One canonical identity across npm, PyPI, OCI, repo, remote & cmd
  • Flags: unclaimed · stale release · no SECURITY.md observed
  • Community reliability & permission profiles (k-anonymized ax-ray signal)
  • Shareable, cacheable report URLs

One model across six identity kinds

bar width = repos observed; (n) = distinct servers

  • npm (1794)2,735 (34%)
  • local command (3273)2,606 (32%)
  • remote endpoint (748)1,281 (16%)
  • PyPI (622)1,130 (14%)
  • OCI / Docker (168)273 (3%)
Market intelligence

Track the MCP ecosystem.

The demand-side view of what the ecosystem actually runs: category leaders, the rising client landscape, and how adoption is distributed — all from public signal anyone can re-derive.

  • Most-adopted servers and the clients driving installs
  • Distribution across identity kinds and co-occurrence clusters
  • Re-derivable, public-signal data — no self-reported numbers
  • Browse the full catalog or dive into ecosystem insights
Demand-side signal from public GitHub configs — repos that opt out of listing are counted but never named. We say observed, measured, listed — never verified or safe. How this is computed. Build on it via the public API, badges & embeds.