Test your RAG pipelines, agents, and MCP clients against known ground truth.
Reproducible datasets and live endpoints for evaluating GenAI systems. No signup, no credentials — open the tool and start testing.
#products
NIAH — needle-in-a-haystack datasets
Generate haystacks with planted needles at known positions, run them through your RAG pipeline or long-context agent, and measure exactly what it recalls. Ground truth included, so scoring is a diff — not a judgment call. Also includes scrape targets for testing web-browsing and extraction agents.
$ open https://niah.heptafox.comMCP Sandbox — zero-auth MCP server
A free, always-on MCP server for testing any MCP client in under a minute. One HTTP endpoint, built-in echo and identity tools, no API keys. Point your client at it and verify the handshake, tool calls, and responses.
$ open https://mcp.heptafox.com#what you can test
- RAG retrieval quality — recall, precision, and long-context degradation against haystacks with known answers.
- MCP client behavior — connection, tool discovery, and tool-call round trips against a deterministic server.
- Regression testing — fixed, versioned test data means a pipeline change shows up as a score change, not a hunch.