MCP-Atlas

MCP-Atlas evaluates agents that work with MCP (Model Context Protocol) servers — the emerging standard for connecting tools. The model must use a realistic set of MCP tools to complete tasks, and an independent judge model checks whether the result is actually correct.

MCP-Atlas measures how well a model acts as a general-purpose agent. In detail, it checks:

• discovering and selecting MCP tools
• correct sequences of tool calls across servers
• reaching the requested end result, verified by a judge
• robustness when tools return errors

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the MCP-Atlas benchmark?

MCP-Atlas evaluates agents that work with MCP (Model Context Protocol) servers — the emerging standard for connecting tools. The model must use a realistic set of MCP tools to complete tasks, and an independent judge model checks whether the result is actually correct.

What does MCP-Atlas test?

MCP-Atlas mainly measures how well a model acts as a general-purpose agent. Concretely, it checks: discovering and selecting MCP tools; correct sequences of tool calls across servers; reaching the requested end result, verified by a judge.

How is the MCP-Atlas score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get MCP-Atlas data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/mcp-atlas/scores) or pull it directly from the Open LLM Radar API with any HTTP client.