Iris

What is Iris?

Iris is an open-source MCP server that logs every trace, evaluates output quality, and tracks costs across all your agents. Any MCP-compatible agent discovers and uses it automatically — no SDK or code changes required.

How to use Iris?

Add Iris to your MCP config (e.g., Claude Desktop) with "command": "npx", "args": ["@iris-eval/mcp-server"]. To enable the web dashboard, run npx @iris-eval/mcp-server --dashboard and open http://localhost:6920. Iris also supports global npm install, Docker, and HTTP transport with optional API key authentication.

Key features of Iris

• Hierarchical trace logging with per-tool costs in USD
• 12 built-in output evaluation rules (completeness, relevance, safety, cost)
• Real-time web dashboard with dark-mode UI
• Cost visibility with aggregate reports and budget thresholds
• Custom eval rules using Zod schemas
• SQLite storage, queryable instantly

Use cases of Iris

• Detect PII leakage or hallucinated answers in agent responses
• Monitor and control per-query costs across multiple agents
• Evaluate agent output quality against custom rules
• Debug inefficient tool-calling patterns in production agents

FAQ from Iris

What runtimes or dependencies does Iris require?

Iris requires Node.js (to run via npx or npm). It also supports Docker deployment.

Where does Iris store its data?

Iris stores logs and evaluations in a local SQLite database, defaulting to ~/.iris/iris.db.

What transport modes does Iris support?

Iris supports both stdio (default) and http transport. The HTTP mode can be secured with an API key, CORS restrictions, and rate limiting.

Is there a cloud version of Iris?

A cloud tier with PostgreSQL, team dashboards, and alerting is coming soon. Self-hosted Iris runs entirely on your machine.

How does Iris handle authentication and security?

When using HTTP transport, Iris supports API key authentication (timing-safe comparison), CORS restricted to localhost by default, rate limiting, Helmet security headers, and Zod input validation.