webmcp/analysis · independent project
About
Mostly I'm just trying to look at all of them without bullshitting myself.
This site publishes the results of an independent, unofficial analysis of the roughly 2,500 submissions to the WebMCP Challenge. It is not an official ranking, not affiliated with the challenge or Devpost, and not a prediction.
Why
A field of 2,500 submissions is too big to reason about from descriptions alone. The analysis reads every entry against the challenge's actual judging criteria, then goes further where evidence allows: demo videos, live products, repositories, and WebMCP runtime verification. The point is a defensible order of the whole field, with the reasoning inspectable rather than hidden behind a single number.
The dataset
- Source: public Devpost submissions to the WebMCP Challenge.
- 2,500 sanitized evidence packets; every project received two blind Stage 1 reviews.
- 716 projects advanced to Stage 2 and were scored with live-product evidence; 1,353 were re-scored with corrected video evidence; 217 have runtime-verified WebMCP behavior.
- The consolidated ranking covers all 2,500 projects.
Caveats
- This is a qualitative analysis by AI reviewers under a frozen protocol, not a ground truth. Read the methodology and the per-project reasoning and judge for yourself.
- Scores rest on the evidence available at each stage; "claims only" means exactly that.
- Authentication is metadata, never a penalty. Pre-existing projects are judged on the submitted concept, not the maturity of the surrounding product.
- An earlier description-first community analysis exists; per protocol it stays quarantined until Stage 4 and does not appear anywhere on this site.
The full protocol, rubric, funnel, and calibration docs live in the analysis repository, alongside every artifact this site displays.