DeepYard

Research

AI Agent Ecosystem Dataset

A reproducible research snapshot of 535 AI agents, MCP servers, frameworks, skills, and developer tools. Every table below is computed from the same listing registry DeepYard uses for directory and editorial pages. Latest listing update: Aug 22, 2026.

Catalog update range: Mar 8, 2026 to Aug 22, 2026. That range reflects the spread of stored listing update dates, not one uniform crawl time.

Total listings

535

Open-source share

482 (90%)

Open-source editorial shortlist

8

Browser-agent editorial shortlist

7

Community signal

Open-source projects by GitHub stars

Read the editorial shortlist

Stars are a discoverability signal, not a quality score. This table includes only open-source listings with a stored public star count and keeps listing freshness visible.

ProjectCategoryStarsListing updated
OpenClawagents387,065Aug 22, 2026
Firecrawldev-tools170,656Aug 22, 2026
Langflowframeworks153,543Aug 22, 2026
Difyframeworks153,166Aug 22, 2026
Open WebUIagents149,532Aug 22, 2026
LangChainframeworks144,745Aug 22, 2026
Awesome LLM Appsagents133,530Aug 22, 2026
Codex CLIagents111,724Aug 22, 2026
Browser Useagents110,043Aug 22, 2026
Gemini CLIagents106,610Aug 22, 2026

Protocol lens

MCP protocol coverage

The directory currently contains 39 MCP listings across 3 stored transport types and 139 distinct capability labels.

Transport distribution

stdio
35 listings
streamable-http
5 listings
sse
3 listings

Most listed compatible clients

claude-code · 37cursor · 24cline · 15codex · 14windsurf · 11continue · 7claude-desktop · 6github-copilot · 3

Category distribution

CategoryCount
agents147
frameworks176
mcp-servers39
skills5
prompts5
workflows21
dev-tools142

Signal coverage

SignalListingsCoverage
GitHub URL14727%
GitHub stars14727%
Contributor count14026%
Last commit14026%

Methodology

  • Counts are computed from the current in-repo listing registry at generation time.
  • Signal coverage counts only reflect the presence of a stored field, not the quality or freshness of that signal beyond the recorded value.
  • Editorial shortlist counts are derived from the same source-of-truth listings and page-specific inclusion rules.
  • Freshness reporting uses listing update ranges and the latest listing update date, because records in the catalog do not all share the same update timestamp.

Limitations

  • This snapshot does not merge private usage data, install telemetry, or purchase data.
  • Some listings are richer than others because upstream projects publish different levels of public documentation and repository metadata.
  • If a listing changes upstream after its stored update date, this snapshot remains stale until the catalog data changes.