# README Score

> Enter a GitHub repo or paste a README: a score out of 100, the sections found and missing, and a fix for each miss. No AI: same README, same score.

## Pages

- [Score a repository](https://readme.flisk.si/flisk-sh/flisk): put the GitHub owner and repository in the address
- [JSON API](https://readme.flisk.si/api/score?repo=flisk-sh/flisk): the same report as JSON; POST {readme} for Markdown
- [MCP server](https://readme.flisk.si/mcp): Streamable HTTP, one read-only tool, score_readme
- [Every check and threshold](https://readme.flisk.si/checks): what each check looks at and its points
- [About and FAQ](https://readme.flisk.si/about): what it reads and stores

## FAQ

- How is the README score worked out? From 28 checks in seven groups: the sections a reader expects for your kind of project (35 points), the first screen (20), readability (10), the quickstart (10), links (10), the tagline and the name's spelling (5), and whether the product's website has the README's headline, pitch, first command and install command (10). A check with nothing to look at is left out, and the score is the points earned over the points that count. Every threshold is listed on the checks page.
- Is AI involved? No. Every check is a fixed rule: a pattern, a count or a formula. The same README always gets the same score, and every finding names the line to change.
- Which repositories can it read? Public GitHub repositories. It reads the README, the file tree, the package manifests, llms.txt files and translated READMEs at one commit, and names that commit on the result. For anything else, paste the README.
- How does it know what kind of project it is? From the files: a bin field in package.json makes a CLI tool, a web framework an app or an API, a main field a library; pyproject.toml, Cargo.toml and go.mod are read the same way. A pasted README is read for signals such as npm install -g or GET /users. You can pick another type on the result page.
- Is there an API or a badge? Yes: /api/score?repo=owner/name answers JSON, /badge/owner/name.svg is a badge with the score, and /mcp is an MCP server with one tool, score_readme, for Claude, ChatGPT, Cursor or Codex. None needs a key.
- What is stored? Nothing on disk. A repository's result is kept in memory for ten minutes, so a shared link does not ask GitHub again, then forgotten. A pasted README is scored and dropped. No cookies.
- Where do the checks come from? From the scoreboard Flisk runs on its own README on every pull request. On the same file, the first-screen, readability and tagline results here are the scoreboard's.

## Optional

- [Source](https://github.com/flisk-sh/readme-score)