A massive empirical study of 177,000 tools reveals a rapid shift in the AI agent ecosystem from 'perception' to 'action' (27% to 65% usage).
March 26, 2026
Original Paper
How are AI agents used? Evidence from 177,000 MCP tools
arXiv · 2603.23802
The Takeaway
This is the first large-scale tracking of the Model Context Protocol (MCP) ecosystem. It provides critical data on the 'consequentiality' of agent tasks, moving the discussion from theoretical agent capabilities to real-world deployment patterns and regulatory needs.
From the abstract
Today's AI agents are built on large language models (LLMs) equipped with tools to access and modify external environments, such as corporate file systems, API-accessible platforms and websites. AI agents offer the promise of automating computer-based tasks across the economy. However, developers, researchers and governments lack an understanding of how AI agents are currently being used, and for what kinds of (consequential) tasks. To address this gap, we evaluated 177,436 agent tools created f