Software is exposing actions directly to AI agents
A protocol and product cohort measures software surfaces designed for agents to execute authenticated actions and transactions.
Early evidence; the market is not yet confirmed
94.4% of the measured cohort is MCP-oriented, while authorization appears in 3.3% and transactions in 4.4%. Agent-facing interfaces are proliferating faster than the trust and transaction layer required for dependable business action.
01
Search-match dynamics
Bars show monthly GitHub total_count plus captured Hugging Face models; the right column shows downloaded items used for detailed analysis.
Jan 26471517 loaded
Feb 26733819 loaded
Mar 26102359 loaded
Apr 26810414 loaded
May 26679010 loaded
Jun 26532013 loaded
Jul 2624738 loaded
02
What exists inside the category
One item may contain more than one feature.
MCP interfaces
85 · 94.4%
A2A and action schemas
1 · 1.1%
Authorization
3 · 3.3%
Audit and approval
4 · 4.4%
Transactions
4 · 4.4%
Business actions
7 · 7.8%
03
Representative projects
githubheadroomlabs-ai/headroom61,132 starsgithubDeusData/codebase-memory-mcp33,961 starsgithubt8y2/dbx11,173 starsgithubgoogle-labs-code/stitch-skills7,759 starsgithubepiral/bb-browser5,988 starsgithubGentleman-Programming/engram5,625 starsgithubTabularisDB/tabularis3,798 starsgithubMouseww/anything-analyzer3,336 starsgithubliaohch3/claude-tap2,570 starsgithubjgravelle/jcodemunch-mcp2,044 stars
04
Cross-source validation
These publications are not part of the primary numeric cohort.