Evaluate MCP agent outputs with standardized quality, safety, and cost scoring.
Copy the install command and let the AI configure it · recommended for beginners
Please install the "io.github.iris-eval/mcp-server" MCP server from askskill: Run: claude mcp add 'io-github-iris-eval-mcp-server' -- npx -y @iris-eval/mcp-server
Please evaluate this MCP customer support agent output using the iris-eval standard. Score quality, safety, and cost, then explain deductions and improvement suggestions: User query: I want a refund Agent output: Hello, please provide your order number and purchase email for the refund, and we will process it as soon as possible.
Returns category scores, an overall assessment, risk notes, and suggestions to improve response quality and efficiency.
Use the iris-eval standard to compare outputs from two MCP agent versions. Focus on answer quality, safety risk, and reasoning cost, then recommend which version is better for release. Version A output: ... Version B output: ...
Provides side-by-side scoring, a strengths and weaknesses summary, and a release recommendation.
Apply the iris-eval standard to this batch of MCP agent evaluation logs. Summarize quality, safety, and cost scores for each output, and identify common issue patterns and outliers.
Outputs a batch scoring summary, issue category statistics, and a list of samples that need deeper review.
Evaluate AI agent outputs for CI gates, regressions, and canary promotions.
Run prompt and RAG evaluations through MCP clients with hosted backend execution.
Test API compatibility for AI agents with scores, grades, and recommendations.
Evaluate AI response quality and behavior patterns during development in real time.
Autonomously evaluate web apps to uncover functionality, performance, and usability issues.
Investigate agents, conversation threads, and content trust signals for analysis.