llms-audit
by F0Rextasy
README.md
# llms-audit — AI-answer readiness checker
**Audit any site for the AI-answer era: llms.txt, schema, sitemap → score + fix list.**
[](https://github.com/F0Rextasy/llms-audit/actions/workflows/test.yml)
[](LICENSE)
[](https://github.com/F0Rextasy/llms-audit/releases)
## Install
```sh
npm i -g github:F0Rextasy/llms-audit
```
Offline-friendly: once fetched, all checks run locally against the downloaded
HTML/robots/sitemap — no account, no API key, no telemetry.
## 30-second setup
```console
$ llms-audit https://yoursite.com
score 5/100 — http://127.0.0.1:8790/
check status fix
-------------- ------ ---
llms.txt FAIL Create /llms.txt: first line `# <site name>`, then at least one `## ` section (run with --fix-llms for a template)
llms-full.txt WARN Publish full context at /llms-full.txt (concatenated markdown of key pages)
sitemap.xml FAIL Publish /sitemap.xml with at least one well-formed <url> entry containing <loc>
robots.txt FAIL Allow GPTBot in robots.txt (remove `Disallow: /` for GPTBot)
JSON-LD FAIL Add <script type="application/ld+json"> with @type Organization/WebSite/WebPage/Article
meta tags FAIL Add missing <title>, meta description, canonical to <head>
<h1> PASS -
HTTPS + health FAIL Serve the site over HTTPS and redirect http → https
```
Fix what it flags, re-run — an AI-ready page (everything present, HTTP local so
only HTTPS fails) scores:
```console
$ llms-audit http://127.0.0.1:8791
score 85/100 — http://127.0.0.1:8791/
check status fix
-------------- ------ ---
llms.txt PASS -
llms-full.txt PASS -
sitemap.xml PASS -
robots.txt PASS -
JSON-LD PASS -
meta tags PASS -
<h1> PASS -
HTTPS + health FAIL Serve the site over HTTPS and redirect http → https
```
Need the llms.txt boilerplate? It prints one for your host:
```console
$ llms-audit https://acme.test --fix-llms
# Acme
> One-paragraph summary of Acme for AI answers. Replace this line with what the site is and who it serves.
## Products
- [Example item](https://acme.test/): one-line description of the key offering.
```
Flags: `--json` (machine report), `--out report.html` (self-contained dark-mode
report you can attach to a PR), `--timeout ms`, `--fix-llms`.
## How it works
Eight weighted checks (100 points total), fetched once from the target origin:
| check | weight | passes when |
|---|---|---|
| llms.txt | 20 | `# site` heading + ≥1 `## ` section |
| llms-full.txt | 5 | full-context file exists (advisory) |
| sitemap.xml | 15 | parses with ≥1 `<url><loc>` |
| robots.txt | 15 | none of GPTBot, ClaudeBot, Claude-Web, Perplexity-Bot, Google-Extended fully disallowed |
| JSON-LD | 15 | parses; `@type` in Organization/WebSite/WebPage/Article |
| meta tags | 10 | `<title>` + meta description + canonical |
| `<h1>` | 5 | exactly one descriptive heading present |
| HTTPS + health | 15 | https final URL, no redirect loop, status < 400 |
Every FAIL/WARN prints the concrete fix, so the report doubles as a worklist.
### MCP server
`llms-audit --mcp` speaks MCP over stdio and exposes `audit_url {url}` — wire it
into Claude Code, Cursor, or any MCP client:
```json
{"mcpServers": {"llms-audit": {"command": "llms-audit", "args": ["--mcp"]}}}
```
## Comparison
| | llms-audit | open-seo-tool style crawlers | manual checklist |
|---|---|---|---|
| Focus | AI-answer readiness (llms.txt, AI bots, JSON-LD) | classic SEO rank factors | you, a notepad |
| Output | score + per-check fix, HTML/SARIF-ready JSON, MCP | dashboards, accounts | inconsistent |
| Cost | free, offline after fetch | subscription | time |
## Regenerate demo
```sh
python tools/render_demo.py # spawns fixtures/server.ts (unready + ready), runs the real CLI, rebuilds assets/
```
Fixtures live in `fixtures/server.ts`; every console block above is a real run.
## License
[MIT](LICENSE)
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessNo issues