Ecosystem Trends

Longitudinal view of the tracked AI-agent ecosystem, computed from 48 daily registry snapshots (since 2026-05-23). Dashed red markers are methodology changes — score-derived series restart there instead of pretending a redefinition is a migration. Quarterly summaries publish as State of Agent Trust reports.

Tracked agents

Actively listed agents in the registry each day.

Listed agents
016232405/2606/2607/26v2.0v3.0v3.1v3.2v4.0v4.2

MCP servers implemented

Agents with public evidence of an implemented MCP server — adoption of the structured tool protocol.

MCP implemented
05310605/2606/2607/26v2.0v3.0v3.1v3.2v4.0v4.2

Trust-grade distribution

Agents per grade band. Lines restart at methodology changes — a cutover redefines the bands, so connecting across one would fake a migration.

Grade AGrade BGrade CGrade D
010120205/2606/2607/26v2.0v3.0v3.1v3.2v4.0v4.2

Provider dependencies

Agents depending on each of the currently-largest runtime providers (detected from manifests and credential markers).

openaianthropicpostgresamazon-bedrockgoogle-geminiredis
013226505/2606/2607/26v2.0v3.0v3.1v3.2v4.0v4.2

Stale agents

Agents with no repository push for 90+ days — the abandonment signal.

Stale ≥90d
0305905/2606/2607/26v2.0v3.0v3.1v3.2v4.0v4.2
Methodology changes in this window: 2026-05-28 → v3.0 · 2026-05-31 → v3.1 · 2026-06-05 → v3.2 · 2026-07-04 → v4.0 · 2026-07-05 → v4.2. Grade bands are redefined at each change (see the methodology changelog), so grade series are deliberately not connected across the markers.