MCP ONBOARDING EXPERIMENTS
A material change. A verified result.
Compare onboarding guides and tool descriptions against a fixed task, then connect feedback to a tested revision.
Explore saved resultsRecorded experiment · synthetic records · local execution. This public explorer makes no model calls.
Inspect the evidence
| Task | Material / trial | Verdict | Evidence | CNY cost bound |
|---|
Read the changed materials
Choose a workflow
The local workspace supports immutable materials, real stdio MCP tools, independent verification, cost reservation and feedback-linked revisions. This website provides the published experiment view.
Interpretation and privacy
The first matrix uses one model, six synthetic task families and correlated repeats. A passed 21/36, B passed 36/36. Guide and descriptions changed together. These results support this comparison; independent developer adoption and market demand await external evidence.
No login, uploads, feedback forms or application analytics are included. Hosting providers may process ordinary access logs. Keys and private execution files stay outside this bundle.