discover-api

Imported from jason920612/local-llm-chat (skills/discover-api/SKILL.md). Install upstream with npx skills add jason920612/local-llm-chat --skill discover-api. Copyright stays with the author.

Discover a site's backend API and call it directly

Many sites render data from a JSON/GraphQL API that's far cleaner to use than scraping HTML. Find that API by inspecting what the site itself does, then request it directly.

Procedure

1. Probe with the bundled script (read-only: tries standard discovery paths, scans the site's own frontend JS for endpoint URLs, checks OpenAPI/Swagger and GraphQL):

   pip install -q requests beautifulsoup4 lxml   # usually already installed
   python scripts/probe.py https://example.com

2. If OpenAPI/Swagger is found (/openapi.json, /swagger.json, /v3/api-docs…): you now have the full endpoint list + parameters — read it and call what you need.

3. If GraphQL is exposed: run an introspection query to get the schema, then write a focused query:

   curl -s -X POST https://example.com/graphql -H 'content-type: application/json' \
     -d '{"query":"{__schema{queryType{name} types{name}}}"}' | python -m json.tool | head -60

4. Otherwise, learn from the frontend: the probe lists /api/... URLs found in the JS. Open the relevant JS (or use the mirror-site skill) and read how it builds the request — method, path, query params, and headers (Content-Type, any public tokens it sends).

5. Test the request directly and inspect the response shape:

   curl -s 'https://example.com/api/v1/items?limit=5' -H 'accept: application/json' | python -m json.tool | head -40
   # or in python: requests.get(url, params=..., headers=...).json()

   Then iterate params until you get the data you need, and use it.

Notes

• Prefer a documented endpoint (OpenAPI/GraphQL schema) over guessing; replicate the exact headers/Content-Type the site's own frontend sends.
• Keep request volume light with a small delay so probing stays fast and reliable; back off on 429.