How are AI agents used? Evidence from 177,000 MCP tools
2026-03-26T07:23:55Z•9b798466224d8fab5147c7eb45c433903adecadc517afc0813b9bd360ac93200
access-controlaction-toolsagent-toolsai-agentsai-ethicsapi-abuseauditing-and-loggingcapability-degradationdata-exfiltrationeducation-aies-llmsexplainabilityfile-system-accessfinancial-fraudhuman-ai-dependencyinterpretabilityknowledge-inequalitymcpopen-source-systemsperception-toolsreasoning-toolsregulation-and-oversighttool-governancetool-supply-chainvirtual-influencers
What happened
This arXiv feed aggregates papers (Mar 2026) about the current landscape and downstream risks of agentic AI and related GenAI systems. The largest study monitored 177,436 Model Context Protocol (MCP) tools (11/2024–02/2026), categorising tools as perception, reasoning, or action; software-development tools dominated (67% of tools, 90% of downloads) and the share of "action" tools rose from 27% to 65% of usage over the sampling window. Most action tools target medium-stakes automation (e.g., file edits), but some enable higher-stakes actions such as agentic financial transactions and direct API
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_cy
- Record identifier
- 9b798466224d8fab5147c7eb45c433903adecadc517afc0813b9bd360ac93200
- Enrichment time
- 2026-03-26T07:23:55Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.