feat: serve the chatbot content manifest at /api/content/manifest

public/typo3conf/ext/ep_products/Classes/ContentFeed/ContentFeedApi.php

@@ -0,0 +1,96 @@
+<?php
+
+namespace EP\EpProducts\ContentFeed;
+
+use EP\EpProducts\Http\ApiKeyAuthenticator;
+use Psr\Http\Message\ResponseInterface;
+use Psr\Http\Message\ServerRequestInterface;
+use TYPO3\CMS\Core\Cache\CacheManager;
+use TYPO3\CMS\Core\Cache\Frontend\FrontendInterface;
+use TYPO3\CMS\Core\Http\JsonResponse;
+
+/**
+ * The chatbot content feed's backchannel endpoints under /api/content/.
+ *
+ * See .ddev/plans/CHATBOT-MCP-01-TYPO3-CONTENT-FEED.md.
+ */
+class ContentFeedApi
+{
+    public const PATH_PREFIX = '/api/content/';
+
+    private const MANIFEST_CACHE_IDENTIFIER = 'content_feed_manifest';
+    // The consumer polls every 15 minutes, so this caps the build at once per poll however
+    // often it retries. Edits reach the chatbot within about 25 minutes.
+    private const MANIFEST_CACHE_LIFETIME = 600;
+
+    private const JSON_FLAGS = JSON_UNESCAPED_UNICODE | JSON_UNESCAPED_SLASHES;
+
+    /**
+     * @var ApiKeyAuthenticator
+     */
+    private $authenticator;
+
+    /**
+     * @var ContentFeedConfiguration
+     */
+    private $configuration;
+
+    /**
+     * @var ManifestBuilder
+     */
+    private $manifestBuilder;
+
+    /**
+     * @var FrontendInterface
+     */
+    private $cache;
+
+    public function __construct(
+        ApiKeyAuthenticator $authenticator,
+        ContentFeedConfiguration $configuration,
+        ManifestBuilder $manifestBuilder,
+        CacheManager $cacheManager
+    ) {
+        $this->authenticator = $authenticator;
+        $this->configuration = $configuration;
+        $this->manifestBuilder = $manifestBuilder;
+        $this->cache = $cacheManager->getCache('ep_products_cache');
+    }
+
+    public function handle(ServerRequestInterface $request): ResponseInterface
+    {
+        if (!$this->authenticator->isAuthorized($request)) {
+            return $this->error('Invalid or missing api key', 403);
+        }
+
+        if ([] === $this->configuration->getSites()) {
+            // An empty feed would read as "everything was deleted". Refuse instead.
+            return $this->error('Content feed has no sites configured', 503);
+        }
+
+        $path = substr($request->getUri()->getPath(), strlen(self::PATH_PREFIX));
+
+        if ('manifest' === $path) {
+            return new JsonResponse($this->getManifest(), 200, [], self::JSON_FLAGS);
+        }
+
+        return $this->error('Not found', 404);
+    }
+
+    private function getManifest(): array
+    {
+        $manifest = $this->cache->get(self::MANIFEST_CACHE_IDENTIFIER);
+
+        if (!is_array($manifest)) {
+            $manifest = $this->manifestBuilder->build();
+            $this->cache->set(self::MANIFEST_CACHE_IDENTIFIER, $manifest, [], self::MANIFEST_CACHE_LIFETIME);
+        }
+
+        return $manifest;
+    }
+
+    private function error(string $message, int $status): ResponseInterface
+    {
+        return new JsonResponse(['message' => $message], $status, [], self::JSON_FLAGS);
+    }
+}

public/typo3conf/ext/ep_products/Classes/ContentFeed/ContentFeedConfiguration.php

@@ -37,6 +37,16 @@ class ContentFeedConfiguration
         return GeneralUtility::trimExplode(',', (string) ($this->configuration['contentFeedSites'] ?? ''), true);
     }
 
+    /**
+     * @return int[] language ids; the default language when unset
+     */
+    public function getLanguages(): array
+    {
+        $languages = GeneralUtility::intExplode(',', (string) ($this->configuration['contentFeedLanguages'] ?? '0'), true);
+
+        return [] === $languages ? [0] : $languages;
+    }
+
     /**
      * @return int[] page uids whose whole subtree stays out of the feed
      */

public/typo3conf/ext/ep_products/Classes/ContentFeed/ManifestBuilder.php

@@ -0,0 +1,351 @@
+<?php
+
+namespace EP\EpProducts\ContentFeed;
+
+use TYPO3\CMS\Core\Database\ConnectionPool;
+use TYPO3\CMS\Core\Site\Entity\Site;
+
+/**
+ * Builds GET /api/content/manifest: everything the content feed publishes, with a modification
+ * marker per item and no bodies. The list is complete - an item missing from it is gone - so the
+ * consumer never has to infer deletions from timestamps.
+ *
+ * See .ddev/plans/CHATBOT-MCP-01-TYPO3-CONTENT-FEED.md §4.
+ */
+class ManifestBuilder
+{
+    public const CONTENT_PAGE_TYPE = 1730;
+
+    private const DOKTYPE_DEFAULT = 1;
+    // Recycler and backend user section: nothing below them is public.
+    private const DOKTYPES_NOT_TRAVERSED = [6, 255];
+    // l18n_cfg bit 1: "hide default translation of page".
+    private const L18N_HIDE_DEFAULT_LANGUAGE = 1;
+
+    private const RECORD_FIELDS = [
+        'product' => ['uid', 'tstamp', 'detail_page'],
+        'hotel' => ['uid', 'tstamp', 'detail_page'],
+        'region' => ['uid', 'tstamp', 'detail_page'],
+        'country' => ['uid', 'tstamp', 'detail_page'],
+        'city' => ['uid', 'tstamp', 'detail_page'],
+        'concept' => ['uid', 'tstamp', 'detail_page'],
+        'news' => ['uid', 'tstamp', 'pid'],
+    ];
+
+    /**
+     * @var PageTree
+     */
+    private $pageTree;
+
+    /**
+     * @var UrlResolver
+     */
+    private $urlResolver;
+
+    /**
+     * @var RecordFinder
+     */
+    private $recordFinder;
+
+    /**
+     * @var ContentFeedConfiguration
+     */
+    private $configuration;
+
+    /**
+     * @var ConnectionPool
+     */
+    private $connectionPool;
+
+    public function __construct(
+        PageTree $pageTree,
+        UrlResolver $urlResolver,
+        RecordFinder $recordFinder,
+        ContentFeedConfiguration $configuration,
+        ConnectionPool $connectionPool
+    ) {
+        $this->pageTree = $pageTree;
+        $this->urlResolver = $urlResolver;
+        $this->recordFinder = $recordFinder;
+        $this->configuration = $configuration;
+        $this->connectionPool = $connectionPool;
+    }
+
+    public function build(): array
+    {
+        $sites = $this->urlResolver->getIncludedSites();
+        $records = $this->buildRecordItems();
+
+        return [
+            'generatedAt' => date('c'),
+            'sites' => array_map([$this, 'describeSite'], $sites),
+            'items' => array_merge($this->buildPageItems($sites, $this->collectDetailPages()), $records),
+        ];
+    }
+
+    private function describeSite(Site $site): array
+    {
+        $languages = [];
+
+        foreach ($site->getLanguages() as $language) {
+            $languages[] = $language->getTwoLetterIsoCode();
+        }
+
+        $title = trim((string) ($site->getConfiguration()['websiteTitle'] ?? ''));
+
+        if ('' === $title) {
+            $title = trim($site->getDefaultLanguage()->getWebsiteTitle());
+        }
+
+        return [
+            'identifier' => $site->getIdentifier(),
+            'title' => '' !== $title ? $title : $site->getIdentifier(),
+            'base' => (string) $site->getBase(),
+            'languages' => $languages,
+        ];
+    }
+
+    /**
+     * @param Site[] $sites
+     * @param array<int, array> $detailPages detailOf entries by page uid
+     */
+    private function buildPageItems(array $sites, array $detailPages): array
+    {
+        $contentTimestamps = $this->findContentTimestamps();
+        $excluded = array_flip($this->configuration->getExcludedPages());
+        $feedLanguages = $this->configuration->getLanguages();
+        $items = [];
+
+        foreach ($sites as $site) {
+            $languageIds = [];
+
+            foreach ($site->getLanguages() as $language) {
+                if (in_array($language->getLanguageId(), $feedLanguages, true)) {
+                    $languageIds[$language->getLanguageId()] = $language->getTwoLetterIsoCode();
+                }
+            }
+
+            $context = [
+                'site' => $site,
+                'languages' => $languageIds,
+                'excluded' => $excluded,
+                'contentTimestamps' => $contentTimestamps,
+                'detailPages' => $detailPages,
+            ];
+            $breadcrumbs = array_fill_keys(array_keys($languageIds), []);
+
+            $this->walk($site->getRootPageId(), $context, $breadcrumbs, true, $items);
+        }
+
+        return $items;
+    }
+
+    /**
+     * @param array<int, string[]> $breadcrumbs titles above this page, by language id
+     */
+    private function walk(int $uid, array $context, array $breadcrumbs, bool $isRoot, array &$items): void
+    {
+        $page = $this->pageTree->getPage($uid);
+
+        if (null === $page || isset($context['excluded'][$uid])) {
+            return;
+        }
+
+        if (in_array((int) $page['doktype'], self::DOKTYPES_NOT_TRAVERSED, true)) {
+            return;
+        }
+
+        $visible = $this->pageTree->isVisible($page);
+
+        if (!$visible && (int) $page['extendToSubpages'] === 1) {
+            return;
+        }
+
+        $childBreadcrumbs = $breadcrumbs;
+
+        foreach ($context['languages'] as $languageId => $isoCode) {
+            $row = $this->getLanguageRow($page, $languageId);
+
+            if ($visible && null !== $row && $this->isPublishable($page, $row, $languageId)) {
+                $item = $this->buildPageItem($page, $row, $languageId, $isoCode, $breadcrumbs[$languageId], $context);
+
+                if (null !== $item) {
+                    $items[] = $item;
+                }
+            }
+
+            if (!$isRoot) {
+                $source = $row ?? $page;
+                $childBreadcrumbs[$languageId][] = '' !== trim((string) $source['nav_title'])
+                    ? trim((string) $source['nav_title'])
+                    : trim((string) $source['title']);
+            }
+        }
+
+        foreach ($this->pageTree->getChildren($uid) as $childUid) {
+            $this->walk($childUid, $context, $childBreadcrumbs, false, $items);
+        }
+    }
+
+    private function getLanguageRow(array $page, int $languageId): ?array
+    {
+        if (0 === $languageId) {
+            return $page;
+        }
+
+        $overlay = $this->pageTree->getOverlay((int) $page['uid'], $languageId);
+
+        return null !== $overlay && $this->pageTree->isVisible($overlay) ? $overlay : null;
+    }
+
+    /**
+     * Mirrors the XML sitemap: standard pages only, nothing marked no_index. If Google is not
+     * meant to see a page, the chatbot is not either.
+     */
+    private function isPublishable(array $page, array $row, int $languageId): bool
+    {
+        if ((int) $page['doktype'] !== self::DOKTYPE_DEFAULT || (int) $row['no_index'] === 1) {
+            return false;
+        }
+
+        return 0 !== $languageId || 0 === ((int) $page['l18n_cfg'] & self::L18N_HIDE_DEFAULT_LANGUAGE);
+    }
+
+    private function buildPageItem(array $page, array $row, int $languageId, string $isoCode, array $breadcrumb, array $context): ?array
+    {
+        $uid = (int) $page['uid'];
+        $url = $this->urlResolver->getPageUrl($uid, $languageId);
+
+        if (null === $url) {
+            return null;
+        }
+
+        $modified = max(
+            (int) $row['tstamp'],
+            $context['contentTimestamps'][$uid][$languageId] ?? 0,
+            $context['contentTimestamps'][$uid][-1] ?? 0
+        );
+
+        return [
+            'type' => 'page',
+            'uid' => $uid,
+            'language' => $isoCode,
+            'site' => $context['site']->getIdentifier(),
+            'url' => $url,
+            'contentUrl' => $this->urlResolver->getPageUrl($uid, $languageId, ['type' => self::CONTENT_PAGE_TYPE]),
+            'title' => trim((string) $row['title']),
+            'description' => $this->nullIfEmpty((string) $row['description']),
+            'breadcrumb' => $breadcrumb,
+            'detailOf' => $context['detailPages'][$uid] ?? [],
+            'modified' => date('c', $modified),
+        ];
+    }
+
+    /**
+     * Latest change to any content element, by page and language. Deleted and hidden elements
+     * count - removing or hiding one changes the page - which is also why SYS_LASTCHANGED is no
+     * use here: it only moves when the page is rendered.
+     *
+     * @return array<int, array<int, int>>
+     */
+    private function findContentTimestamps(): array
+    {
+        $queryBuilder = $this->connectionPool->getQueryBuilderForTable('tt_content');
+        $queryBuilder->getRestrictions()->removeAll();
+
+        $rows = $queryBuilder
+            ->selectLiteral('pid', 'sys_language_uid', 'MAX(tstamp) AS tstamp')
+            ->from('tt_content')
+            ->groupBy('pid', 'sys_language_uid')
+            ->execute()
+            ->fetchAllAssociative()
+        ;
+
+        $timestamps = [];
+
+        foreach ($rows as $row) {
+            $timestamps[(int) $row['pid']][(int) $row['sys_language_uid']] = (int) $row['tstamp'];
+        }
+
+        return $timestamps;
+    }
+
+    /**
+     * Which records name a page as their detail page. Keyed by the page the record points at,
+     * not by where a shortcut leads: a shortcut is not in the manifest, and the page it leads to
+     * usually describes something broader (product detail shortcuts point at region pages).
+     *
+     * @return array<int, array>
+     */
+    private function collectDetailPages(): array
+    {
+        $detailPages = [];
+
+        foreach (self::RECORD_FIELDS as $type => $fields) {
+            if (!in_array('detail_page', $fields, true)) {
+                continue;
+            }
+
+            foreach ($this->recordFinder->find($type, ['uid', 'detail_page']) as $uid => $row) {
+                $pageUid = $this->urlResolver->parsePageReference((string) $row['detail_page']);
+
+                if (null !== $pageUid) {
+                    $detailPages[$pageUid][] = ['type' => $type, 'uid' => $uid];
+                }
+            }
+        }
+
+        return $detailPages;
+    }
+
+    private function buildRecordItems(): array
+    {
+        $faqTimestamps = $this->recordFinder->findFaqTimestamps();
+        $matchcodeTimestamps = $this->recordFinder->findMatchcodeTimestamps();
+        $items = [];
+
+        foreach (self::RECORD_FIELDS as $type => $fields) {
+            foreach ($this->recordFinder->find($type, $fields) as $uid => $row) {
+                if ('news' === $type) {
+                    $link = $this->urlResolver->getNewsLink((int) $row['uid'], (int) $row['pid']);
+
+                    // News has no detail_page of its own: an article stored outside the feed's
+                    // sites (the retired ep-events) is not part of the feed at all.
+                    if (null === $link) {
+                        continue;
+                    }
+                } else {
+                    $link = $this->urlResolver->getRecordLink((string) $row['detail_page']);
+                }
+
+                $modified = (int) $row['tstamp'];
+
+                if ('product' === $type) {
+                    $modified = max($modified, $faqTimestamps[$uid] ?? 0);
+                }
+
+                if ('hotel' === $type) {
+                    $modified = max($modified, $matchcodeTimestamps[$uid] ?? 0);
+                }
+
+                $items[] = [
+                    'type' => $type,
+                    'uid' => $uid,
+                    'language' => 'de',
+                    'site' => $link['site'],
+                    'url' => $link['url'],
+                    'modified' => date('c', $modified),
+                ];
+            }
+        }
+
+        return $items;
+    }
+
+    private function nullIfEmpty(string $value): ?string
+    {
+        $value = trim($value);
+
+        return '' === $value ? null : $value;
+    }
+}

public/typo3conf/ext/ep_products/Classes/ContentFeed/PageTree.php

@@ -0,0 +1,131 @@
+<?php
+
+namespace EP\EpProducts\ContentFeed;
+
+use TYPO3\CMS\Core\Context\Context;
+use TYPO3\CMS\Core\Database\ConnectionPool;
+use TYPO3\CMS\Core\Database\Query\Restriction\DeletedRestriction;
+
+/**
+ * The whole page table in memory, for the content feed's tree walk and shortcut resolution.
+ *
+ * About 2 900 rows: one query and a PHP walk is far cheaper than a query per tree level.
+ * Visibility is evaluated here, not in SQL, because a hidden page still has to be traversed
+ * (its subpages stay reachable unless extendToSubpages is set).
+ */
+class PageTree
+{
+    private const FIELDS = [
+        'uid', 'pid', 'sorting', 'doktype', 'hidden', 'starttime', 'endtime', 'fe_group',
+        'extendToSubpages', 'no_index', 'title', 'nav_title', 'description', 'shortcut',
+        'shortcut_mode', 'tstamp', 'sys_language_uid', 'l10n_parent', 'l18n_cfg',
+    ];
+
+    /**
+     * @var ConnectionPool
+     */
+    private $connectionPool;
+
+    /**
+     * @var int
+     */
+    private $now;
+
+    /**
+     * @var array<int, array>|null default-language rows by uid
+     */
+    private $pages;
+
+    /**
+     * @var array<int, int[]> child uids by parent uid, in sorting order
+     */
+    private $children = [];
+
+    /**
+     * @var array<int, array<int, array>> overlay rows by default-language uid and language id
+     */
+    private $overlays = [];
+
+    public function __construct(ConnectionPool $connectionPool, Context $context)
+    {
+        $this->connectionPool = $connectionPool;
+        $this->now = (int) $context->getPropertyFromAspect('date', 'timestamp');
+    }
+
+    public function getPage(int $uid): ?array
+    {
+        $this->load();
+
+        return $this->pages[$uid] ?? null;
+    }
+
+    public function getOverlay(int $uid, int $languageId): ?array
+    {
+        $this->load();
+
+        return $this->overlays[$uid][$languageId] ?? null;
+    }
+
+    /**
+     * @return int[]
+     */
+    public function getChildren(int $uid): array
+    {
+        $this->load();
+
+        return $this->children[$uid] ?? [];
+    }
+
+    /**
+     * Visible to an anonymous frontend visitor: not hidden, inside its start/end window, and not
+     * restricted to a frontend user group.
+     */
+    public function isVisible(array $row): bool
+    {
+        if ((int) $row['hidden'] === 1) {
+            return false;
+        }
+
+        if ((int) $row['starttime'] > $this->now) {
+            return false;
+        }
+
+        if ((int) $row['endtime'] !== 0 && (int) $row['endtime'] <= $this->now) {
+            return false;
+        }
+
+        return in_array((string) $row['fe_group'], ['', '0'], true);
+    }
+
+    private function load(): void
+    {
+        if (null !== $this->pages) {
+            return;
+        }
+
+        $queryBuilder = $this->connectionPool->getQueryBuilderForTable('pages');
+        $queryBuilder->getRestrictions()->removeAll()->add(new DeletedRestriction());
+
+        $rows = $queryBuilder
+            ->select(...self::FIELDS)
+            ->from('pages')
+            ->orderBy('sorting')
+            ->execute()
+            ->fetchAllAssociative()
+        ;
+
+        $this->pages = [];
+
+        foreach ($rows as $row) {
+            $uid = (int) $row['uid'];
+
+            if ((int) $row['sys_language_uid'] > 0) {
+                $this->overlays[(int) $row['l10n_parent']][(int) $row['sys_language_uid']] = $row;
+                continue;
+            }
+
+            $this->pages[$uid] = $row;
+            $this->children[(int) $row['pid']][] = $uid;
+        }
+    }
+}

public/typo3conf/ext/ep_products/Classes/ContentFeed/RecordFinder.php

@@ -0,0 +1,152 @@
+<?php
+
+namespace EP\EpProducts\ContentFeed;
+
+use TYPO3\CMS\Core\Database\Connection;
+use TYPO3\CMS\Core\Database\ConnectionPool;
+use TYPO3\CMS\Core\Database\Query\QueryBuilder;
+
+/**
+ * Which records the content feed publishes. The manifest and the records endpoint both go
+ * through here, so a uid the manifest lists is exactly a uid the records endpoint returns.
+ *
+ * The default restrictions apply: deleted, hidden, starttime and endtime.
+ */
+class RecordFinder
+{
+    public const TYPES = [
+        'product' => 'tx_epproducts_domain_model_product',
+        'hotel' => 'tx_epproducts_domain_model_hotel',
+        'region' => 'tx_epproducts_domain_model_region',
+        'country' => 'tx_epproducts_domain_model_country',
+        'city' => 'tx_epproducts_domain_model_city',
+        'concept' => 'tx_epproducts_domain_model_concept',
+        'news' => 'tx_news_domain_model_news',
+    ];
+
+    /**
+     * @var ConnectionPool
+     */
+    private $connectionPool;
+
+    public function __construct(ConnectionPool $connectionPool)
+    {
+        $this->connectionPool = $connectionPool;
+    }
+
+    /**
+     * @param string[] $fields
+     * @param int[]|null $uids null for all eligible records
+     * @return array<int, array> rows by uid, in uid order
+     */
+    public function find(string $type, array $fields, ?array $uids = null): array
+    {
+        $table = self::TYPES[$type];
+        $queryBuilder = $this->connectionPool->getQueryBuilderForTable($table);
+        $queryBuilder
+            ->select(...$fields)
+            ->from($table)
+            ->where($queryBuilder->expr()->in('sys_language_uid', [0, -1]))
+            ->orderBy('uid')
+        ;
+
+        if (null !== $uids) {
+            if ([] === $uids) {
+                return [];
+            }
+
+            $queryBuilder->andWhere($queryBuilder->expr()->in(
+                'uid',
+                $queryBuilder->createNamedParameter($uids, Connection::PARAM_INT_ARRAY)
+            ));
+        }
+
+        $this->addTypeConstraints($type, $queryBuilder);
+
+        $rows = [];
+
+        foreach ($queryBuilder->execute()->fetchAllAssociative() as $row) {
+            $rows[(int) $row['uid']] = $row;
+        }
+
+        return $rows;
+    }
+
+    /**
+     * Latest change to a product's FAQs, by product uid. Deleted and hidden FAQs count: removing
+     * one is a change to the product's text.
+     *
+     * @return array<int, int>
+     */
+    public function findFaqTimestamps(): array
+    {
+        $queryBuilder = $this->connectionPool->getQueryBuilderForTable('tx_epproducts_domain_model_faq');
+        $queryBuilder->getRestrictions()->removeAll();
+
+        $rows = $queryBuilder
+            ->selectLiteral('mm.uid_local AS product', 'MAX(faq.tstamp) AS tstamp')
+            ->from('tx_epproducts_product_faq_mm', 'mm')
+            ->join('mm', 'tx_epproducts_domain_model_faq', 'faq', 'faq.uid = mm.uid_foreign')
+            ->groupBy('mm.uid_local')
+            ->execute()
+            ->fetchAllAssociative()
+        ;
+
+        return $this->indexTimestamps($rows, 'product');
+    }
+
+    /**
+     * Latest change to a hotel's matchcode aliases, by hotel uid. They are part of the hotel's
+     * `codes`, which the consumer joins its trip data on.
+     *
+     * @return array<int, int>
+     */
+    public function findMatchcodeTimestamps(): array
+    {
+        $queryBuilder = $this->connectionPool->getQueryBuilderForTable('tx_epproducts_domain_model_matchcode');
+        $queryBuilder->getRestrictions()->removeAll();
+
+        $rows = $queryBuilder
+            ->selectLiteral('entity AS hotel', 'MAX(tstamp) AS tstamp')
+            ->from('tx_epproducts_domain_model_matchcode')
+            ->where($queryBuilder->expr()->eq('entity_type', $queryBuilder->createNamedParameter('hotel')))
+            ->groupBy('entity')
+            ->execute()
+            ->fetchAllAssociative()
+        ;
+
+        return $this->indexTimestamps($rows, 'hotel');
+    }
+
+    /**
+     * @return array<int, int>
+     */
+    private function indexTimestamps(array $rows, string $key): array
+    {
+        $timestamps = [];
+
+        foreach ($rows as $row) {
+            $timestamps[(int) $row[$key]] = (int) $row['tstamp'];
+        }
+
+        return $timestamps;
+    }
+
+    private function addTypeConstraints(string $type, QueryBuilder $queryBuilder): void
+    {
+        $expr = $queryBuilder->expr();
+
+        if ('product' === $type) {
+            // A product that is neither searchable nor bookable is an internal record.
+            $queryBuilder->andWhere($expr->orX($expr->eq('searchable', 1), $expr->eq('bookable', 1)));
+        }
+
+        if ('news' === $type) {
+            // Type 0 is a regular article; 1 and 2 are just links to a page or an external URL.
+            $queryBuilder->andWhere(
+                $expr->eq('type', $queryBuilder->createNamedParameter('0')),
+                $expr->in('fe_group', $queryBuilder->createNamedParameter(['', '0'], Connection::PARAM_STR_ARRAY))
+            );
+        }
+    }
+}

public/typo3conf/ext/ep_products/Classes/ContentFeed/UrlResolver.php

@@ -0,0 +1,214 @@
+<?php
+
+namespace EP\EpProducts\ContentFeed;
+
+use TYPO3\CMS\Core\Exception\SiteNotFoundException;
+use TYPO3\CMS\Core\Site\Entity\Site;
+use TYPO3\CMS\Core\Site\SiteFinder;
+use TYPO3\CMS\Core\Utility\MathUtility;
+
+/**
+ * Absolute public URLs for the content feed, generated through each site's router so that
+ * baseVariants apply (staging and DDEV get their own hosts) and route enhancers shape the path.
+ */
+class UrlResolver
+{
+    private const DOKTYPE_DEFAULT = 1;
+    private const DOKTYPE_SHORTCUT = 4;
+    private const SHORTCUT_MODE_NONE = 0;
+    private const SHORTCUT_MODE_FIRST_SUBPAGE = 1;
+    private const SHORTCUT_MODE_PARENT_PAGE = 3;
+    private const MAX_SHORTCUT_DEPTH = 5;
+
+    /**
+     * @var SiteFinder
+     */
+    private $siteFinder;
+
+    /**
+     * @var PageTree
+     */
+    private $pageTree;
+
+    /**
+     * @var string[]
+     */
+    private $includedSites;
+
+    /**
+     * @var array<string, int> news detail page uid by site identifier
+     */
+    private $newsDetailPages;
+
+    /**
+     * @var array<int, Site|null>
+     */
+    private $sitesByPage = [];
+
+    public function __construct(SiteFinder $siteFinder, PageTree $pageTree, ContentFeedConfiguration $configuration)
+    {
+        $this->siteFinder = $siteFinder;
+        $this->pageTree = $pageTree;
+        $this->includedSites = $configuration->getSites();
+        $this->newsDetailPages = $configuration->getNewsDetailPages();
+    }
+
+    /**
+     * @return Site[] included sites, in configuration order
+     */
+    public function getIncludedSites(): array
+    {
+        $sites = [];
+
+        foreach ($this->includedSites as $identifier) {
+            try {
+                $sites[] = $this->siteFinder->getSiteByIdentifier($identifier);
+            } catch (SiteNotFoundException $e) {
+                // A misspelt identifier must not take the whole feed down; it just contributes nothing.
+            }
+        }
+
+        return $sites;
+    }
+
+    /**
+     * The page's site, or null when it belongs to no site or to one outside the feed.
+     */
+    public function getIncludedSite(int $pageUid): ?Site
+    {
+        if (!array_key_exists($pageUid, $this->sitesByPage)) {
+            try {
+                $site = $this->siteFinder->getSiteByPageId($pageUid);
+                $this->sitesByPage[$pageUid] = in_array($site->getIdentifier(), $this->includedSites, true) ? $site : null;
+            } catch (SiteNotFoundException $e) {
+                $this->sitesByPage[$pageUid] = null;
+            }
+        }
+
+        return $this->sitesByPage[$pageUid];
+    }
+
+    public function getPageUrl(int $pageUid, int $languageId = 0, array $parameters = []): ?string
+    {
+        $site = $this->getIncludedSite($pageUid);
+
+        if (null === $site) {
+            return null;
+        }
+
+        try {
+            $language = $site->getLanguageById($languageId);
+
+            return (string) $site->getRouter()->generateUri($pageUid, ['_language' => $language] + $parameters);
+        } catch (\Throwable $e) {
+            // Unroutable pages (missing slug, disabled language) are left out rather than failing the feed.
+            return null;
+        }
+    }
+
+    /**
+     * Parses a typolink page reference as stored in `detail_page`: a plain uid or `t3://page?uid=N`.
+     */
+    public function parsePageReference(string $reference): ?int
+    {
+        $reference = trim($reference);
+
+        if (MathUtility::canBeInterpretedAsInteger($reference)) {
+            return (int) $reference > 0 ? (int) $reference : null;
+        }
+
+        if (1 === preg_match('/^t3:\/\/page\?(?:.*&)?uid=(\d+)/', html_entity_decode($reference), $matches)) {
+            return (int) $matches[1];
+        }
+
+        return null;
+    }
+
+    /**
+     * Follows shortcut pages the way a typolink does, and returns the uid of the visible standard
+     * page a link to $pageUid ends up on - or null when the chain breaks or loops.
+     */
+    public function resolveLinkTarget(int $pageUid): ?int
+    {
+        for ($depth = 0; $depth < self::MAX_SHORTCUT_DEPTH; $depth++) {
+            $page = $this->pageTree->getPage($pageUid);
+
+            if (null === $page || !$this->pageTree->isVisible($page)) {
+                return null;
+            }
+
+            if ((int) $page['doktype'] !== self::DOKTYPE_SHORTCUT) {
+                return (int) $page['doktype'] === self::DOKTYPE_DEFAULT ? $pageUid : null;
+            }
+
+            $next = $this->getShortcutTarget($page);
+
+            if (null === $next) {
+                return null;
+            }
+
+            $pageUid = $next;
+        }
+
+        return null;
+    }
+
+    /**
+     * @return array{url: ?string, site: ?string}
+     */
+    public function getRecordLink(string $detailPage): array
+    {
+        $pageUid = $this->parsePageReference($detailPage);
+        $target = null === $pageUid ? null : $this->resolveLinkTarget($pageUid);
+        $site = null === $target ? null : $this->getIncludedSite($target);
+        $url = null === $site ? null : $this->getPageUrl($target);
+
+        return [
+            'url' => $url,
+            'site' => null === $url ? null : $site->getIdentifier(),
+        ];
+    }
+
+    /**
+     * News has no detail_page of its own: the storage folder decides the site, and the site's
+     * configured detail page carries the article through the NewsDetailPlugin route enhancer.
+     *
+     * @return array{url: ?string, site: string}|null null when the article is stored outside the feed's sites
+     */
+    public function getNewsLink(int $newsUid, int $storagePid): ?array
+    {
+        $site = $this->getIncludedSite($storagePid);
+
+        if (null === $site) {
+            return null;
+        }
+
+        $detailPage = $this->newsDetailPages[$site->getIdentifier()] ?? null;
+        $url = null === $detailPage ? null : $this->getPageUrl($detailPage, 0, [
+            'tx_news_pi1' => ['action' => 'detail', 'controller' => 'News', 'news' => $newsUid],
+        ]);
+
+        return ['url' => $url, 'site' => $site->getIdentifier()];
+    }
+
+    private function getShortcutTarget(array $page): ?int
+    {
+        switch ((int) $page['shortcut_mode']) {
+            case self::SHORTCUT_MODE_NONE:
+                return (int) $page['shortcut'] > 0 ? (int) $page['shortcut'] : null;
+            case self::SHORTCUT_MODE_FIRST_SUBPAGE:
+                foreach ($this->pageTree->getChildren((int) $page['uid']) as $childUid) {
+                    if ($this->pageTree->isVisible($this->pageTree->getPage($childUid))) {
+                        return $childUid;
+                    }
+                }
+
+                return null;
+            case self::SHORTCUT_MODE_PARENT_PAGE:
+                return (int) $page['pid'] > 0 ? (int) $page['pid'] : null;
+            default:
+                // Random subpage: there is no stable URL to give.
+                return null;
+        }
+    }
+}

public/typo3conf/ext/ep_products/Classes/Middleware/ApiMiddleware.php

@@ -2,6 +2,7 @@
 
 namespace EP\EpProducts\Middleware;
 
+use EP\EpProducts\ContentFeed\ContentFeedApi;
 use EP\EpProducts\Service\MyEpImageService;
 use EP\EpProducts\Utility\HotelIconUtility;
 use Psr\Http\Message\ResponseInterface;
@@ -41,18 +42,25 @@ class ApiMiddleware implements MiddlewareInterface
      */
     private $imageService;
 
+    /**
+     * @var ContentFeedApi
+     */
+    private $contentFeedApi;
+
     public function __construct(
         MyEpImageService $hotelImageService,
         MyEpImageService $productImageService,
         MyEpImageService $regionImageService,
         FileRepository $fileRepository,
-        ImageService $imageService
+        ImageService $imageService,
+        ContentFeedApi $contentFeedApi
     ) {
         $this->hotelImageService = $hotelImageService;
         $this->productImageService = $productImageService;
         $this->regionImageService = $regionImageService;
         $this->fileRepository = $fileRepository;
         $this->imageService = $imageService;
+        $this->contentFeedApi = $contentFeedApi;
     }
 
     public function process(ServerRequestInterface $request, RequestHandlerInterface $handler): ResponseInterface
@@ -67,6 +75,10 @@ class ApiMiddleware implements MiddlewareInterface
             return $this->handleProductRequest($request);
         }
 
+        if (0 === strpos($path, ContentFeedApi::PATH_PREFIX)) {
+            return $this->contentFeedApi->handle($request);
+        }
+
         return $handler->handle($request);
     }
 

public/typo3conf/ext/ep_products/ext_conf_template.txt

@@ -85,6 +85,9 @@ contentFeedEnabled = 1
 # cat=ContentFeed; type=string; label=Site identifiers included in the chatbot content feed (comma-separated)
 contentFeedSites = ep-reisen,ser,ski-boarderweek,ski-wochenenden,snowzone
 
+# cat=ContentFeed; type=string; label=Language ids included in the chatbot content feed (comma-separated; the English pages are untranslated placeholders today)
+contentFeedLanguages = 0
+
 # cat=ContentFeed; type=string; label=Page uids whose subtrees stay out of the chatbot content feed (comma-separated; 634 = group booking links)
 contentFeedExcludedPages = 634