io.github.iris-eval/mcp-serverai

Iris

Verified · 6 days ago

Stop shipping agents on vibes. Score every agent output for quality, safety, and cost.

Install

claude mcp add mcp-server -- npx -y @iris-eval/[email protected]

Requires: IRIS_API_KEY, IRIS_API_KEY_FILE, IRIS_ALLOW_UNAUTHENTICATED, IRIS_DB_PATH, IRIS_SQLITE_DRIVER, IRIS_LOG_LEVEL, IRIS_HOME, IRIS_ANTHROPIC_API_KEY, IRIS_OPENAI_API_KEY, IRIS_LLM_JUDGE_MAX_COST_USD_PER_EVAL, IRIS_CITATION_ALLOW_FETCH, IRIS_DASHBOARD, IRIS_DASHBOARD_PORT, IRIS_PORT, IRIS_OTEL_ENDPOINT, IRIS_CITATION_DOMAINS, IRIS_TRANSPORT, IRIS_HOST, IRIS_ALLOWED_ORIGINS, IRIS_DASHBOARD_HOST, IRIS_OTEL_HEADERS, IRIS_OTEL_SERVICE_NAME, IRIS_OTEL_TIMEOUT_MS, IRIS_WEBHOOK_URL, IRIS_WEBHOOK_SECRET

Our take

This server is an evaluation layer for MCP-based agents, aimed at scoring output quality, safety, and cost and enforcing budget or policy checks. It is for teams building or operating agents who need repeatable QA and governance, not for end-user chat workflows. Quality signal looks promising: the repo is substantial and includes tests, docs, examples, dashboard code, and security material, but the category is still young and buyers should expect some ecosystem churn around eval standards.

reviewed by hand · 2026-07-12

Something wrong or dead here? Report it