Open research assets
MCP research you can inspect and cite
Every benchmark below comes from the same 520-listing MCPtrove snapshot. Each page publishes its method, limitations, and the underlying CSV rows so the headline can be checked instead of merely repeated.
MCP Authentication Statistics 2026: API Keys Still Outnumber OAuth 4 to 1
245 listings use no server-specific credential, 222 use an API key, and only 53 use OAuth.
MCP Transport Statistics 2026: 89% of Servers Still Run on stdio
463 of 520 directory listings use stdio, versus 47 on HTTP and 10 on legacy SSE.
MCP Server Language Statistics 2026: TypeScript and Python Build 79% of the Ecosystem
TypeScript and Python account for 408 of 520 server listings, or 78.5% of the directory snapshot.
MCP Tool Count Benchmark 2026: The Median Server Exposes 10 Tools
The median server exposes 10 tools, the average is 13.2, and the 90th percentile is 26.
MCP Server Category Statistics 2026: Developer Tools Lead, but the Market Is Fragmented
116 of 520 listings are developer tools, while the other 404 span 11 distinct categories.
MCP GitHub Stars Benchmark 2026: The Median Listed Server Has 250 Stars
The median listing has 250 GitHub stars, while the 90th percentile has 3,335.
MCP Maintenance Activity Statistics 2026: 59% Had a Commit Within 180 Days
305 of 516 listings with a known commit date had activity within 180 days of September 23, 2026.
MCP Server License Statistics 2026: MIT and Apache Cover 85% of Listings
MIT and Apache-2.0 together cover 442 of 520 listings, or 85.0% of the snapshot.
MCP Installation Method Statistics 2026: npx Leads 40% of Listings
209 listings use npx, 99 use uvx, and 56 use a direct remote URL as their normalized primary method.
MCP Official and Verified Statistics 2026: Only 16% Carry Both Signals
85 of 520 listings are both official and verified, while 331 carry neither signal.
What this dataset can and cannot answer
These are descriptive statistics for listings normalized by MCPtrove, not a census of every server ever published. They are useful for comparing implementation patterns and choosing a smaller, safer tool stack. For individual recommendations, browse servers by capability or run your config through the Config Doctor.