MCP is a common protocol for connecting AI applications to tools and data. In the 2025 report, its significance was adoption across major platforms, reducing one-off integrations while creating new security responsibilities.
These are separate benchmarks with different tasks and environments. Their percentages should not be averaged or interpreted as a universal probability of an agent completing office work. The report identifies continuing long-horizon failures.
Evidence you can use
Results reported in the 2025 edition
| Measure | Reported value | Definition and source |
|---|---|---|
| OSWorld | 47.5% | Benchmark score2025 report, slide 81 (PDF page 82) |
| WindowsAgentArena | 50.6% | Benchmark score2025 report, slide 81 (PDF page 82) |
| AndroidWorld | 73.3% | Benchmark score2025 report, slide 81 (PDF page 82) |
| Online-Mind2Web | 88.2% | Benchmark score2025 report, slide 81 (PDF page 82) |
These are separate benchmarks with different tasks and environments. Their percentages should not be averaged or interpreted as a universal probability of an agent completing office work. The report identifies continuing long-horizon failures.
Historical snapshot published October 9, 2025. This web edition was prepared on October 10, 2026 from the online deck and original launch posts. Findings and forecasts retain their original time frame.
Benaich, Nathan. “What is Model Context Protocol?.” State of AI Report 2025. Historical report snapshot; web edition prepared 2026-10-10.