ReviewIQ: practice spotting code mistakes, with Zod-validated tools

[Anas127]

Hi, I'm Anas, the founder of ReviewIQ. It is practice for the step after learning to code: read unfamiliar code, explain the mistakes, then compare your findings with caught/missed feedback. It is not a course or a credential.

One implementation detail for Zod users: the MCP endpoint uses Zod at the tool boundary, while the application grades the meaning of the written findings separately. For example, the current submit_review tool declares:

inputSchema: {
  userReview: z.string().trim().min(1).max(5000),
}

A nonempty string with a valid length is only an acceptable input shape; it is not evidence that the learner identified a real problem. Keeping those two checks separate also keeps validation errors distinct from feedback about the review itself.

The actual implementation is in app/mcp/route.ts. The project declares Zod ^4.6.5.

The website has one fixed Backend/Python/Mid-level guest exercise, free before signup. An account is needed to save and continue. Try the exercise.

Below is an actual founder test, not customer proof: 4.3/10, 1 issue caught and 2 missed.

For others building learning tools: which checks do you keep in the input schema, and which do you reserve for the feedback layer?