See what your AI agents
are actually doing.
Iris is an open-source MCP server for agent trace logging, quality evaluation, and observability. Works with any MCP-compatible framework.
Trace Logging
Log every agent execution with spans, tool calls, token usage, and cost. Structured traces with parent-child span relationships, just like OpenTelemetry.
Quality Evaluation
Score agent outputs against 13 built-in rules for completeness, relevance, safety, and cost. Add custom rules with regex, keywords, or JSON schema validation.
Observability Dashboard
Dark-mode web dashboard with real-time trace viewer, span trees, evaluation results, and summary charts. Self-hosted, no cloud dependency.
Up and running in 60 seconds
One command to install. One command to start.
# Install and start
npx @iris-eval/mcp-server --dashboard
# Or add to Claude Desktop config
{ "mcpServers": { "iris": { "command": "npx",
"args": ["@iris-eval/mcp-server"] } } }
npx @iris-eval/mcp-server --dashboard
# Or add to Claude Desktop config
{ "mcpServers": { "iris": { "command": "npx",
"args": ["@iris-eval/mcp-server"] } } }
Built for developers
Works with any MCP-compatible agent
Claude Desktop
Cursor
LangChain
CrewAI
AutoGen
Custom Agents