NazarKalytiuk/tarn

What it does

Tarn is an API testing tool designed specifically for AI agent integration. It outputs structured JSON failure responses with categorized error information (failure_category, error_code, hints) rather than unstructured stderr output. This allows AI agents to programmatically understand and branch on different failure types without complex text parsing.

Key features

• CLI-first design with MCP server integration
• Structured JSON failure responses with taxonomy categories
• Multiple MCP tools: tarn_run, tarn_validate, tarn_fix_plan, tarn_inspect, tarn_rerun_failed
• YAML-based test configuration (.tarn.yaml files)
• Cross-platform support (macOS, Windows, Linux)
• AI-agent optimized error reporting with hints and error codes

How to set up

Tarn is available as an MCP server and can be integrated into Claude, ChatGPT, or other AI platforms supporting the MCP protocol. Installation involves adding the server configuration to your MCP settings and defining your API tests in .tarn.yaml files following the tool's specification format.