Software development dominates AI agent tool usage

Researchers analysed 177,436 AI agent tools created between November 2024 and February 2026 using Model Context Protocol repositories. The study examines how AI agents use external tools to access and modify digital environments.

The tools are grouped into perception, reasoning and action categories based on their function. Perception tools access data, reasoning tools analyse information, and action tools modify systems such as files, emails or external platforms.

Software development accounts for 67% of all tools and 90% of downloads. The findings show that AI agents are primarily used to support coding tasks and related workflows.

The share of action tools increased from 27% to 65% over the 16 months analysed. Most action tools focus on medium-stakes tasks, though some are used for financial transactions and other higher-stakes activities.

The study also outlines a method to monitor AI agent usage through tool-level analysis. This approach can support oversight of risks linked to AI deployment in practical applications.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!