MCP-Universe
Agents · 2025-08-20
MCP-Universe evaluates agents on real-world tasks served through Model Context Protocol servers across location navigation, repository management, financial analysis, 3D design, browser automation and web search, with execution-based evaluators. Scores are task success rates; the server environment and agent loop are run-level settings and some reports use their own environments. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Pro Preview | 50.7 |
| GPT-5 | 47.9 |
| Sonnet 4.5 | 46.5 |
| DeepSeek-V3.2 | 45.9 |
| Kimi K2 Thinking | 35.6 |
| MiniMax M2 | 29.4 |