feat: llms.txt and llms-full.txt, openapi.json, MCP server

docs/llm/gotchas.md

@@ -0,0 +1,67 @@
+# Gotchas
+
+> What trips up API clients: session-only routes, error texts, rate limits and draft visibility.
+
+## Session-only routes
+
+An API key cannot call these routes. They answer `403` with a plain-text body and no JSON:
+
+```
+HTTP/1.1 403 Forbidden
+
+This action needs a real session.
+```
+
+- `/v1/api-key/*`, `/v1/system/*`, `POST /v1/auth/verify`, `DELETE /v1/auth/register`.
+- Every activity route under `/v1/activity`.
+- Writes on sites (`POST`, `PATCH`, `DELETE` under `/v1/site`, so a key cannot create a site) and on authors (`POST`, `PATCH`, `DELETE` under `/v1/author`).
+- `PATCH /v1/settings`.
+
+Variant routes under `/v1/media/{id}/variants` answer `401` to a key instead. The sign-in routes (login, logout, register, password reset) ignore a key and give no `403`.
+
+When a task needs one of these routes, stop and tell the user to do it in the app. Do not try to log in with a password.
+
+## Error texts are plain text
+
+Errors are not JSON. The body is one sentence; branch on the status code, and show the sentence to the user.
+
+| Status | Typical text |
+| --- | --- |
+| `400` | `The request cannot be processed due to a mismatch in the format of the body.` |
+| `401` | `You must be logged in to access this resource.` |
+| `403` | `This action needs a real session.`, `Only the owner can do this.`, `Only owners and admins can do this.`, `You must verify your account to access this resource.` |
+| `404` | `Article not found`, `Image not found` |
+| `429` | `Too many requests. Please try again later.` |
+
+An invalid, unknown, expired or revoked key answers `401`, even on public routes. It is never treated as anonymous.
+
+## Rate limit: 429
+
+Each key may make 120 requests per minute. The `429` response carries a `Retry-After` header holding the seconds to wait:
+
+```
+HTTP/1.1 429 Too Many Requests
+Retry-After: 46
+
+Too many requests. Please try again later.
+```
+
+Sleep for `Retry-After` seconds, then retry once. Do not retry in a tight loop. Prefer one list call with `size=100` over many single reads.
+
+## Draft versus online
+
+- Anonymous callers and unverified users only see online articles. A draft answers `404` with `Article not found`, the same as an id that does not exist.
+- A verified session or a key sees drafts as well. In lists, `counts.draft` is `0` for anonymous callers and `status=draft` answers an empty page.
+- `POST /v1/article/` always creates a draft. Publishing is a separate call, `PATCH /v1/article/{id}/publish`.
+- Sitemaps list online articles only.
+
+## Writes
+
+- A title cannot be changed after creation. Create a new article instead.
+- `PATCH /v1/article/{id}` accepts `body`, `site_id`, `author_ids` and the SEO fields only. `{}` and `null` values answer `400`, except `cover_image_id` and `social_image_id`, where `null` removes the image.
+- Request bodies are HTML even though responses default to markdown. Send `<p>` and `<h2>`, not markdown syntax, or it is stored literally.
+- Ids in paths are unsigned integers. Anything else answers `400`.
+
+## Permissions follow the key's creator
+
+A key acts as the user who made it. A key made by a user who is neither owner nor admin cannot do what only owners and admins can, and it answers `403 Only owners and admins can do this.` Check `GET /v1/auth/me` to read the role of the key.

docs/llm/recipe-articles.md

@@ -0,0 +1,81 @@
+# Recipe: create, edit and publish an article
+
+> Create a draft, replace its body, check its SEO, publish it and read it back, with an API key.
+
+Set the key once. Every request below is real output from a running instance; ids and timestamps differ on yours.
+
+```sh
+export RENEWCMS_API_KEY=rcms_...
+```
+
+## 1. Create a draft
+
+Titles are 3 to 200 characters and cannot be changed later. The request body is HTML; the response body is markdown unless you pass `?format=html` or `?format=json`.
+
+```sh
+curl -X POST "{{BASE_URL}}/v1/article/" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"title":"How we price a bathroom renovation","body":"<p>A bathroom renovation is priced by room size, plumbing changes and finish level.</p>"}'
+```
+
+Status `201`:
+
+```json
+{"id":1,"title":"How we price a bathroom renovation","body":"A bathroom renovation is priced by room size, plumbing changes and finish level.","body_format":"markdown","is_online":false,"site_id":1,"authors":[],"created_at":"2026-10-23T01:11:16.767761797+02:00","updated_at":"2026-10-23T01:11:16.767761797+02:00","slug":"how-we-price-a-bathroom-renovation","excerpt":"","seo_title":"","seo_description":"","cover_image_id":null,"cover_image_alt":"","social_title":"","social_description":"","social_image_id":null,"canonical_url":"","noindex":false,"seo_dismissed":[]}
+```
+
+The article starts offline (`is_online: false`) on the default site.
+
+## 2. Replace the body, move it to a site, set the excerpt
+
+`PATCH` takes only `body`, `site_id`, `author_ids` and the SEO fields. At least one is required; `{}` answers `400`. Send `{"body": ""}` to empty a body.
+
+```sh
+curl -X PATCH "{{BASE_URL}}/v1/article/1" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"body":"<h2>What drives the price</h2><p>Room size, plumbing changes and finish level.</p>","site_id":2,"excerpt":"What drives the price of a bathroom renovation."}'
+```
+
+Status `200`:
+
+```json
+{"id":1,"title":"How we price a bathroom renovation","body":"## What drives the price\n\nRoom size, plumbing changes and finish level.","body_format":"markdown","is_online":false,"site_id":2,"authors":[],"created_at":"2026-10-23T01:11:16.767761797+02:00","updated_at":"2026-10-23T01:11:16.771008847+02:00","slug":"how-we-price-a-bathroom-renovation","excerpt":"What drives the price of a bathroom renovation.","seo_title":"","seo_description":"","cover_image_id":null,"cover_image_alt":"","social_title":"","social_description":"","social_image_id":null,"canonical_url":"","noindex":false,"seo_dismissed":[]}
+```
+
+## 3. Check the SEO before publishing
+
+An empty JSON object checks the saved article. Pass any SEO field to check a change without saving it.
+
+```sh
+curl -X POST "{{BASE_URL}}/v1/article/1/seo/check" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{}'
+```
+
+Status `200`:
+
+```json
+{"status":"suggestions","suggestions":[{"code":"desc_short","field":"seo_description","params":{"length":47,"min":70}},{"code":"cover_none","field":"cover_image_id","params":{}},{"code":"no_au","field":"author_ids","params":{}},{"code":"short","field":"body","params":{"length":66,"min":300}}],"slug":{"available":true,"suggestion":null}}
+```
+
+Suggestions never block publishing.
+
+## 4. Publish
+
+```sh
+curl -X PATCH "{{BASE_URL}}/v1/article/1/publish" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+Status `200`, same article object with `"is_online":true`. `PATCH /v1/article/{id}/unpublish` reverses it.
+
+## 5. Read it back without credentials
+
+```sh
+curl "{{BASE_URL}}/v1/article/1?format=markdown"
+```
+
+Before step 4 this answers `404` with the text `Article not found`, because anonymous callers only see online articles.

docs/llm/recipe-media.md

@@ -0,0 +1,62 @@
+# Recipe: upload an image and attach it to an article
+
+> Upload a file with multipart form data, find it in the library and use it as an article cover.
+
+## 1. Upload
+
+Send the file in a form field named `file`. PNG, JPEG, WEBP and SVG are accepted.
+
+```sh
+curl -X POST "{{BASE_URL}}/v1/media/" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -F "file=@north-wall.png"
+```
+
+Status `201`:
+
+```json
+{"id":1,"filename":"north-wall.png","url":"/static/uploadedImages/8d9be79c-5db6-4d1d-9f06-b5d1b15cb4f0.png","size":100,"content_type":"image/png","width":32,"height":32,"sha256":"7654144c6e1ee8939fa061fa8129763869fa3c2de55954053a53a965848ef09c","uploaded_by":"florian","created_at":"2026-10-23T01:11:24.228947139+02:00","used_in":0,"duplicate":false,"variant_count":0,"variant_size":0}
+```
+
+`url` is relative to the instance: the public address is `{{BASE_URL}}` followed by `url`. `duplicate: true` means the same bytes were already in the library.
+
+## 2. Find an image
+
+```sh
+curl "{{BASE_URL}}/v1/media/?size=5" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+Status `200`:
+
+```json
+{"items":[{"id":1,"filename":"north-wall.png","url":"/static/uploadedImages/8d9be79c-5db6-4d1d-9f06-b5d1b15cb4f0.png","size":100,"content_type":"image/png","width":32,"height":32,"sha256":"7654144c6e1ee8939fa061fa8129763869fa3c2de55954053a53a965848ef09c","uploaded_by":"florian","created_at":"2026-10-23T01:11:24.228947139+02:00","used_in":0,"duplicate":false,"variant_count":0,"variant_size":0}],"total":1,"page":1,"size":5,"counts":{"all":1,"unused":1,"duplicates":0}}
+```
+
+Filter with `q`, `hash`, `filter` (`all`, `unused`, `duplicates`), `sort` and `variants`.
+
+## 3. Attach it as the cover
+
+An image is attached by id. `cover_image_alt` is the alternative text.
+
+```sh
+curl -X PATCH "{{BASE_URL}}/v1/article/1" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"cover_image_id":1,"cover_image_alt":"A tiled bathroom wall"}'
+```
+
+Status `200`, the article object with `"cover_image_id":1` and `"cover_image_alt":"A tiled bathroom wall"`. Use `social_image_id` for the sharing image, and `null` to remove either image.
+
+## 4. See where an image is used
+
+```sh
+curl "{{BASE_URL}}/v1/media/1/usage" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+```json
+[{"type":"article","id":1,"title":"How we price a bathroom renovation","published":true,"fields":["cover"]}]
+```
+
+Deleting or merging an image that is in use changes those articles, so read this first.

docs/llm/recipe-pagination.md

@@ -0,0 +1,67 @@
+# Recipe: paginate, filter and choose a body format
+
+> List articles page by page, search and sort them, and read bodies as markdown, HTML or JSON blocks.
+
+`GET /v1/article/` always answers an envelope, never a bare array:
+
+```sh
+curl "{{BASE_URL}}/v1/article/?format=html&size=5&page=1" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+```json
+{"items":[{"id":1,"title":"How we price a bathroom renovation","body":"<h2>What drives the price</h2><p>Room size, plumbing changes and finish level.</p>","body_format":"html","is_online":true,"site_id":2,"authors":[],"created_at":"2026-10-23T01:11:16.767761797+02:00","updated_at":"2026-10-23T01:11:24.236138156+02:00","excerpt":"What drives the price of a bathroom renovation.","slug":"how-we-price-a-bathroom-renovation","seo_status":"suggestions","seo_suggestions":3}],"total":1,"page":1,"size":5,"counts":{"all":1,"online":1,"draft":0,"suggestions":1}}
+```
+
+In JSON, `<` and `>` are escaped as `<` and `>`; a JSON parser returns the plain characters.
+
+## Walking every page
+
+1. Request `page=1` with the `size` you want (`5`, `25`, `50` or `100`).
+2. Read `total` and `size`. The number of pages is `ceil(total / size)`.
+3. Request the next pages until `page * size >= total`. A page past the end answers `200` with `items: []`, so an empty `items` also ends the loop.
+
+Ordering is applied to the whole list before slicing and ends with the id, so pages never overlap or skip.
+
+## Filters
+
+Combine them freely.
+
+| Parameter | Example | Effect |
+| --- | --- | --- |
+| `q` | `q=grout` | Title contains the text, ignoring case and accents. Up to 200 characters. |
+| `status` | `status=draft` | `online` or `draft`. |
+| `site` | `site=renew-blog` | Only that site's articles. |
+| `seo` | `seo=suggestions` | `ready`, `suggestions` or `hidden`. |
+| `sort` | `sort=title:asc` | `title`, `status`, `created_at` or `updated_at`, then `:asc` or `:desc`. Default `updated_at:desc`. |
+
+`counts` ignores every filter, so tab totals do not move while searching. `total` follows the filters.
+
+```sh
+curl "{{BASE_URL}}/v1/article/?site=renew-blog&status=draft&sort=title:asc&q=grout" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+```json
+{"items":[{"id":2,"title":"Draft about grout","body":"x","body_format":"markdown","is_online":false,"site_id":2,"authors":[],"created_at":"2026-10-23T01:12:24.922895943+02:00","updated_at":"2026-10-23T01:12:24.926728869+02:00","excerpt":"","slug":"draft-about-grout","seo_status":"suggestions","seo_suggestions":4}],"total":1,"page":1,"size":25,"counts":{"all":2,"online":1,"draft":1,"suggestions":2}}
+```
+
+## The `format` parameter
+
+| Value | `body` | `body_format` |
+| --- | --- | --- |
+| absent or `markdown` | markdown string | `markdown` |
+| `html` | the stored HTML | `html` |
+| `json` | array of blocks | `json` |
+
+It applies to every article route, including the writes. Request bodies of `POST` and `PATCH` stay HTML whatever the format. An invalid `format` is reported before any other error.
+
+## Invalid values
+
+A bad `page`, `size`, `status`, `sort` or `seo` answers `400` with a plain-text body:
+
+```
+The request cannot be processed due to a mismatch in the format of the body.
+```
+
+Media lists paginate the same way (`GET /v1/media/?page=2&size=50`) with their own `counts`.

docs/llm/recipe-sites.md

@@ -0,0 +1,66 @@
+# Recipe: sites and sitemaps
+
+> Create a site (session only), move an article onto it and read its public sitemap.
+
+An article belongs to one site. A site has a `key` (a slug used in URLs) and a `base_url` (the public address its pages are served from).
+
+## 1. List the sites
+
+```sh
+curl "{{BASE_URL}}/v1/site/" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY"
+```
+
+```json
+[{"id":1,"name":"Default","key":"default","base_url":"http://localhost:18080","default_social_image_id":null,"default_robots":"index","locale":null,"is_default":true,"article_count":0,"site_name":"","title_template":"%title% | %site%","default_description":""}]
+```
+
+## 2. Create a site: not with a key
+
+Creating, editing, defaulting and deleting a site need a browser session from the owner. A key answers `403`:
+
+```sh
+curl -X POST "{{BASE_URL}}/v1/site/" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"x","key":"x","base_url":"https://x.example"}'
+```
+
+```
+HTTP/1.1 403 Forbidden
+
+This action needs a real session.
+```
+
+Ask the owner to create the site in the app, or send the request with the owner's `jwt` cookie. With a session the same request answers `201`:
+
+```json
+{"id":2,"name":"Renew blog","key":"renew-blog","base_url":"https://blog.renew.example","default_social_image_id":null,"default_robots":"index","locale":null,"is_default":false,"article_count":0,"site_name":"","title_template":"%title% | %site%","default_description":""}
+```
+
+## 3. Put an article on the site
+
+Use the site `id` as `site_id`:
+
+```sh
+curl -X PATCH "{{BASE_URL}}/v1/article/1" \
+  -H "Authorization: Bearer $RENEWCMS_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"site_id":2}'
+```
+
+## 4. Read the sitemap
+
+The sitemap is public and lists the online articles of one site, by its `key`:
+
+```sh
+curl "{{BASE_URL}}/v1/site/renew-blog/sitemap"
+```
+
+```json
+[{"url":"https://blog.renew.example/how-we-price-a-bathroom-renovation","modified_at":"2026-10-22T23:11:24Z","noindex":false}]
+```
+
+`url` is the site's `base_url` plus the article slug. Drafts are not listed. The response header `X-Total-Count` holds the number of entries. Page with `page` and `size` (`25`, `50`, `100` or `500`, default `100`); any other `size` answers `400`. An unknown key answers `404` with `site_not_found`.
+
+List only one site's articles with `GET /v1/article/?site=renew-blog`.