Skip to main content
Data report

State of AI Agents 2026

A snapshot of the AI agents ecosystem, derived directly from the HeyClaude registry. Agents act with real autonomy for Claude and coding workflows — this report covers what they take on, how consistently they disclose their safety and privacy behavior, and what they need to run.

Data as of June 20, 2026 (UTC).

106
Total agents
registry
87
Document safety & privacy
87%
100
Source-backed
100%
37
Ready to use
37%

Most common agent use cases

The work agents take on, from their registry tags (mechanism tags like “agents” excluded). An agent can cover several use cases.

agents-md
14
13%
mcp
12
11%
testing
11
10%
automation
8
8%
enterprise
8
8%
performance
8
8%
architecture
7
7%
security
7
7%
governance
6
6%
react
6
6%

Safety & privacy disclosure

Agents act with real autonomy, so disclosure matters. How many document both their safety and privacy behavior, one, or neither.

Safety & privacy
92
87%
Safety only
1
1%
Privacy only
6
6%
Neither documented
7
7%

Setup prerequisites

Whether an agent needs prerequisites (accounts, tools, or config) before it runs.

Requires prerequisites
67
63%
No prerequisites
39
37%

Download the data

Every figure in this report is available as a machine-readable export, regenerated from the registry. Free to reuse under CC BY 4.0 with attribution.

Methodology & citation

Figures are computed at build time from the 106 agents in the HeyClaude registry, snapshot dated June 20, 2026. Use cases come from registry tags (mechanism tags such as “agents” are excluded); safety and privacy disclosure reflects whether each agent documents that behavior; setup prerequisites and source provenance are recorded during registry review.

Citing this report? Link to heyclau.de/state-of-ai-agents with the data-as-of date. See also the State of Agent Skills and the broader State of Claude Tooling. Browse all agents.

Track the Claude Code ecosystem

A weekly digest of new agents, skills, and MCP servers as they land in the registry.