Who it is for

The intended user already has a draft in Word, Markdown, or LaTeX, plus the data tables and papers behind that draft. Adjacent fields such as nursing, public health, or psychology can use the same workflow. It is not built for someone who only has a topic and wants an AI to produce a first manuscript.

What to prepare

  • The draft you are actually submitting or revising.
  • The data table behind the numbers, if the manuscript reports results from a table.
  • The reference list you cited, and any PDFs you already have locally.
  • A Windows 10 or 11 x64 computer. Account activation, AI review, and online literature lookup need a network connection.

Keep those files in one project folder. MediCrab treats that folder as the project; you do not upload the folder to the website.

What it can check

When you open the folder, a local first pass can already flag some deterministic problems—for example an abstract that says 410 participants while the results say 412. That kind of mismatch is illustrated on the homepage with simulated figures, not a real patient cohort.

A full-manuscript check then collects numbers, citations, and consistency findings into one list. Citation identity is resolved against source records (including PubMed, Crossref, and Chinese DOI services where they apply). Findings are separated so that verified items, items that need attention, missing records, sources the databases do not cover, and failed requests are not collapsed into a single “error” state.

You can also ask for a second model opinion on one conclusion. Disagreement is a reason for you to look, not a proof that either answer is correct.

How to run it

  • Install MediCrab on the machine that already holds the project and activate an account in the app.
  • Choose the project folder, then import the draft you are revising. The original file stays unchanged.
  • Run the full-manuscript check and open the list of findings.
  • Read each finding next to its evidence. Change the manuscript only where you agree.

Online model review and literature lookup consume compute credits. First-time activation includes promotional credits; the app does not promise a fixed number of full reviews.

What still needs a person

A finding is a reason to review the evidence, not a certificate. “Not found” does not mean the citation is fake, and an uncovered source does not mean the sentence is wrong. Failed requests are a separate outcome from missing records.

Original files remain read-only. Suggested edits become changes only after you confirm them. MediCrab does not replace scientific judgment, statistical review, or journal guidelines.

Whether a cited paper actually supports a specific claim depends on the text you have. If a local full text is available, MediCrab can extract passages for review; that is not a substitute for reading the paper, and coverage is limited when the full text is not in the project.