let MDN search ask MDN

MDN renders its search results client-side, so 'oc mdn search' has gone
through DuckDuckGo with a site: filter. But the page it renders gets its
results from a public JSON endpoint, /api/v1/search, which returns the
site's own ranking with a summary per result in one ~11KB response. A new
'api' shortcut shape names such an endpoint and which response fields
hold the list, title, URL, and snippet; the response becomes the same
synthetic results page a Sphinx search produces and rides the normal
render path, so 'do <n>' follows a result. First view is ~520 tokens and
works for any site whose search answers as JSON.

README.md

@@ -126,12 +126,12 @@ Works on any mostly-static site with no per-site setup: news sites, blogs, docum
 | Google Cloud docs | `oc gcp` (via docs.cloud.google.com, search via DuckDuckGo) | `docs <product>`, `page <product> <page>`, `gcloud <command>`, `search <query>` |
 | Microsoft Learn | `oc learn` (search via its RSS API) | `azure <page>`, `doc <path>`, `cli <command>`, `search <query>` |
 | Python docs | `oc py` (search via the docs' own index) | `library <module>`, `doc <path>`, `search <query>` |
-| MDN | `oc mdn` (search via DuckDuckGo) | `js <page>`, `css <page>`, `doc <path>`, `search <query>` |
+| MDN | `oc mdn` (search via the site's own API) | `js <page>`, `css <page>`, `doc <path>`, `search <query>` |
 | Node.js docs | `oc node` (search via DuckDuckGo) | `api <module>`, `search <query>` |
 
 A shortcut only ever resolves to a URL and then takes the same path `oc open` does, so it changes nothing about what a page costs or how it reads. The last argument takes every word after it, so `oc ddg search claude code cli` and `oc aws search s3 lifecycle rules` need no quoting, and a path argument keeps its slashes, so `oc learn doc azure/aks/what-is-aks` reaches that page.
 
-A few of these (X, Stack Overflow, YouTube, Microsoft Learn search) read pages that look login-gated or JS-only from the outside, by finding the server-rendered HTML, feed, inline data, or public API the page already ships without a login. Stack Overflow search goes through the Stack Exchange API, and each result prints its `question_id`: read one with the `question <id>` feed rather than following its link, since the question page itself answers a bot challenge instead of the question. AWS, Google Cloud, MDN, and Node.js render docs search client-side or ship none at all, so their `search` goes through DuckDuckGo with a baked-in `site:` filter instead. Python's docs are built with Sphinx, which publishes the site's full-text search index as one static file, so `oc py search` fetches that index (cached on disk for a day), ranks it locally, and prints a numbered result list; a query that names a symbol exactly, like `json.dumps`, links straight to its anchor. The same backend will work for any Sphinx site, including most Read the Docs projects. Not supported yet: pages that only render with JavaScript, sites behind logins, and sites with hard bot challenges that expose no feed.
+A few of these (X, Stack Overflow, YouTube, Microsoft Learn search) read pages that look login-gated or JS-only from the outside, by finding the server-rendered HTML, feed, inline data, or public API the page already ships without a login. Stack Overflow search goes through the Stack Exchange API, and each result prints its `question_id`: read one with the `question <id>` feed rather than following its link, since the question page itself answers a bot challenge instead of the question. AWS, Google Cloud, and Node.js render docs search client-side or ship none at all, so their `search` goes through DuckDuckGo with a baked-in `site:` filter instead. Python's docs are built with Sphinx, which publishes the site's full-text search index as one static file, so `oc py search` fetches that index (cached on disk for a day), ranks it locally, and prints a numbered result list; a query that names a symbol exactly, like `json.dumps`, links straight to its anchor. The same backend will work for any Sphinx site, including most Read the Docs projects. MDN also renders its search client-side, but the page gets its results from a public JSON endpoint, so `oc mdn search` asks that endpoint directly and prints the site's own ranking; that `api` shape in a site definition works for any site whose search answers as JSON. Not supported yet: pages that only render with JavaScript, sites behind logins, and sites with hard bot challenges that expose no feed.
 
 Want a website on that list? Open a pull request, or an issue naming the site; see [CONTRIBUTING.md](CONTRIBUTING.md).