Ecosystem data
updated just now

The MCP and Agent Skill ecosystem, in numbers.

A live census of every published MCP server and agent skill in the MCPMarkets catalog. Figures are computed on demand — use Refresh for the latest reading. Methodology and daily snapshots are below.

13,815published entries
6,723MCP servers
7,092Agent skills
217added in the last 24h

As of Oct 18, 2026, MCPMarkets tracks 13,815 published MCP servers and agent skills.

Median stars
57MCP servers
138Agent skills
Half of all entries sit below this line.
Top 1% star share
49.3%MCP servers
43.3%Agent skills
Server-side Gini 0.870
Active in 6 months
64.6%MCP servers
86.6%Agent skills
At least one push in the last six months.
License unclear
22.4%MCP servers
30.4%Agent skills
Unlabeled plus GitHub NOASSERTION.
Distribution

The long tail, measured

How GitHub stars are spread across 6,723 MCP servers.

Median57
Mean484
P90532
P997,811
Max96,534

As of Oct 18, 2026, 66.7% of MCP server repositories have fewer than 100 stars; the median is 57.

Concentration

A very narrow head

The Lorenz curve shows how much of all stars the largest repositories hold.

0%50%100%
Repositories, cumulative shareStars, cumulative share
Top 10 share17.9%
Top 1% share49.3%
Gini0.870
Largest entries

The biggest repositories

Ranked by GitHub stars. Classification follows our ingestion heuristics.

  1. 1punkpeye/awesome-mcp-servers ★ 96,534MIT
  2. 2headroomlabs-ai/headroom ★ 75,611Apache-2.0
  3. 3sansan0/trendradar ★ 62,975GPL-3.0
  4. 4zylon-ai/private-gpt ★ 57,594Apache-2.0
  5. 5chromedevtools/chrome-devtools-mcp ★ 53,748Apache-2.0

The top 1% of MCP server repositories hold 49.3% of all stars (Gini 0.870).

Licensing

What can you actually reuse?

License mix across both catalogs. “Unclear” is the bucket that makes commercial use a question.

MCP servers

Permissive 72.7% Copyleft 4.8% Unlabeled 16.6% Unrecognized 5.8%
MIT 58.1%Apache-2.0 13.2%AGPL-3.0 2.1%GPL-3.0 1.7%GPL-2.0 0.5%ISC 0.4%BSD-3-Clause 0.4%MPL-2.0 0.4%

Among MCP server entries, 22.4% have no detectable license, which leaves commercial use ambiguous.

Agent skills

Permissive 66.7% Copyleft 2.9% Unlabeled 22.5% Unrecognized 7.9%
MIT 53.7%Apache-2.0 11.4%AGPL-3.0 1.4%GPL-3.0 1%CC0-1.0 0.5%BSD-3-Clause 0.3%CC-BY-4.0 0.3%MIT-0 0.2%

Among agent skill entries, 30.4% have no detectable license, which leaves commercial use ambiguous.

Maintenance

Is it still alive?

Time since the last push to the source repository.

MCP serversAgent skills
≤ 6 months
64.6%
86.6%
6–12 months
14.9%
13.3%
1–2 years
20.4%
0.1%
> 2 years
0%
0%
No data
0.1%
0.1%

64.6% of MCP server repositories have been pushed to in the last six months.

0% of MCP server repositories have been inactive for more than two years.

Methodology

How these numbers are produced

  • Source: the MCPMarkets catalog, built from GitHub's public API (stars, license, last push, README).
  • Snapshot: Oct 18, 2026 (UTC day). Historical dates replay the payload stored at the end of that day.
  • Counting: published entries only. Total = MCP servers + agent skills; duplicates across the two catalogs are not merged.
  • Freshness: statistics are computed when the page loads and cached for 60 seconds; Refresh recomputes immediately. The catalog itself is refreshed by the crawler and the nightly metrics sync. A JSON snapshot is stored daily.
  • License groups: permissive = MIT/Apache/BSD/ISC/CC0/Unlicense; copyleft = GPL/AGPL/LGPL/MPL/EPL; unclear = missing value or GitHub NOASSERTION.
  • Stars measure attention, not usage or quality. The server/skill split is heuristic and may misclassify a few list-style repositories.
  • Coverage: a snapshot can lag the catalog by a few minutes while the crawler publishes new entries.
History

Snapshot archive

Every stored UTC day can be cited as a fixed URL.

Cite this page

Published under CC BY 4.0. This line is ready to paste into an article or report.

MCPMarkets, “MCP and Agent Skill Ecosystem Insights,” Oct 18, 2026. https://mcpmarkets.com/en/insights