Official scripts
Explore our curated collection of scripts
Submit URL to Brave
Submit one URL to Brave Search to be re-crawled, and return the outcome Brave gave: whether it was accepted, plus Brave's own heading and message. Acceptance means Brave took the submission — not that the page was crawled, indexed or ranked. An invalid URL comes back as a rejection with Brave's message rather than as an error. Brave sometimes shows a rate-limit challenge here; if it does, the script raises an error — wait 30-60 seconds before retrying.
Brave search
Search the web on Brave, one page of results at a time. Returns each result's title, url, snippet and date, plus the query it actually ran. An empty result list always means Brave itself reported nothing for the query, and the reason is returned alongside it — the script raises an error rather than quietly returning nothing when the page cannot be read, so an empty answer can be trusted as a fact about the query. Search operators (site:, quoted phrases, OR) are honored. site:domain also covers subdomains, and site:domain/path narrows to a path prefix. When too few pages match, Brave drops the operators and answers a looser query instead; the result says so, and those results should not be treated as filtered. For coverage questions, site:domain on its own is a ranked sample rather than a full inventory — add search terms, and judge by whether real articles come back or just homepages and section pages. Paging deeper does not enumerate a domain either: the pool behind a single query is finite and Brave re-serves earlier pages once it runs out, so dedupe results by url, stop when a page adds nothing new, and widen coverage with more query strings rather than more pages. Brave sometimes shows a rate-limit challenge; the script clears it automatically, and if it persists it raises an error — wait 30-60 seconds before retrying.
Search Make.com public templates
Search Make's public template catalogue by name and/or by the apps a template uses.
Search Dribbble shots
Search Dribbble shots by keyword, returning title, url, owner, thumbnail, likes and views for each result.
Search ComeUp freelancers
Search ComeUp sellers/freelancers by keyword. Returns username, display name, verified badge and avatar for each match (ComeUp's seller-search results are minimal — use get_profile with the returned username for headline, bio, stats, portfolio and services).
Search Indeed jobs
Search Indeed's public job listings by keyword, optional free-text location and country site (e.g. "fr" for fr.indeed.com, default "us"). Returns total, page, outOfCountry, and jobs (jobKey, title, company, location, salary, jobTypes, remote, sponsored, postedAgo, url). Anonymous search only reliably supports page=1 — Indeed gates further pages behind a sign-in wall, and the script throws loudly if that gate appears. total/results are only meaningful for the requested location when outOfCountry is false; indeed.com (US) treats non-US locations as out-of-country and returns unrelated jobs.
Get Behance project
Fetch a Behance project's details from its gallery URL: title, description, creators, tags, images, likes and views.
Search Behance projects
Search Behance projects by keyword, returning title, url, owner, thumbnail, likes and views for each result.
Search Luma events by topic or city
Search upcoming public events on Luma, filtered by a category topic (e.g. "crypto", "fitness") or a curated city (e.g. "paris", "london"), paginated.
Search Built In jobs
Search Built In (builtin.com) tech/startup job listings by free-text keyword and optional free-text location (city). Returns id, title, url, company, companyUrl, postedAgo, remoteType (Remote/Hybrid/In-Office wording as shown), locations (array — cards with several offices list each one), salary (raw site text, e.g. "180K-201K Annually" — not split into min/max since the site itself only shows a range string, sometimes absent when undisclosed), seniority, industries (array), description, topSkills (array). Paginates via the site's own page param (page=1 default, 25 results/page); the reported last page reflects the true total, not just the current page's count. A query/location with no matches returns jobs:[] — not an error.
Search Claude chats
Search your Claude.ai chats (the "Search chats…" box): server-side full-text over titles + message bodies. Each hit carries the matched snippet and match offsets the UI bolds.
Search Gemini chats
Search your Gemini conversations (the "Search chats" box): server-side full-text over titles and message bodies.
Get YC company profile
Get one YC company's full public profile (ycombinator.com/companies/<slug>): pitch, batch/status, founders, socials, open jobs, news, launches, photos, founder Q&A.
Get Loom video transcript
Fetch the full transcript of a Loom video from its share URL (or bare 32-char id). Returns title, owner, duration, language, timestamped segments, and the joined plain text. Works anonymously on link-shared videos; password-protected or private videos fail loudly, and videos without a transcript return an empty segments list.
List Lovable projects
List all your Lovable projects. Returns name, URL, thumbnail, last edited date, and owner.
Get LinkedIn newsletter edition comments
Get the comments of a LinkedIn newsletter edition by permalink, loading up to maxComments top-level threads and each commenter's profile. Returns total and comments (urn, name, headline, profileUrl, postedAgo, text, reactions, nested replies, hasMoreReplies). The default 'relevant' sort hides low-engagement comments - use 'recent' to reach every comment; older replies behind 'See previous replies' stay unloaded (flagged by hasMoreReplies).
Ask Le Chat
Open a new Le Chat conversation, send one question, wait for the answer, and return the answer text plus any web-search source domains cited.
Get LinkedIn profile comments
Get the comments a LinkedIn profile has left on posts, from its Comments activity feed by public ID, scrolling to load up to count comments. Returns each comment's text/urn/age/reactions and the post it was made on (author, text, url).
instagram.comLike Instagram post
Like an Instagram post or reel by shortcode. Clicks the action-bar heart and confirms the Like→Unlike flip (server-driven). Fails loudly if the post is already liked (no Like button) or unavailable. Returns shortcode and liked.
Neon login (Google SSO)
Log into the Neon console via Google OAuth (through Neon's Keycloak): clicks Google and lands on your Neon projects. Skips straight to your projects if you're already signed in.
Search Slack messages
Full-text search across the workspace's messages, supporting Slack's modifiers (from:@user, in:#channel, before:/after:date, "exact phrase"). One page per call. Requires login.
Search Jooble jobs
Searches job listings on jooble.org (global/US, English UI) by driving the site's own search form (keyword + city) and reading the results it returns. Dismisses the cookie-consent banner if present. Pagination beyond the first page is loaded via progressive scroll, matching the site's own infinite-scroll behavior. One thing to be aware of: Jooble is sensitive to request rate and can temporarily block rapid repeated searches within a short window; if a search fails, space out retries rather than repeating it immediately.
Get Pinterest pin details
Fetches a single Pinterest pin's public details by id: title, description, image, creator, save/reaction counts, publish date, and (for link pins) the linked destination site. Read-only, no login required.
Search Free-Work jobs
Searches job listings on free-work.com (IT/freelance), usable without logging in. `query` is a free-text keyword; `location` is a free-text city, resolved to the closest match — unresolved locations throw explicitly rather than silently searching everywhere. Defaults to searching all of France when no location is given. Returns title, full description, company, contract types, experience level, remote work, salaries (daily/annual), currency, skills, publish date, and URL. 1-based pagination, 16 results per page. One thing to check: requesting a page beyond the real number of pages doesn't return an empty result — it silently repeats the first page — so verify against `totalItems` before trusting a high page number.
List connector Claude directory
Every connector in Claude's Directory (Settings → Connectors → Add → Browse connectors) — the whole catalogue the page loads, not one page. Each entry carries id, slug, name, one-liner, verified tier (anthropic/partner/community), categories, author, its claude.ai/directory link and the MCP endpoint URL. The Directory's Plugins tab (local extensions) is a different affordance and is not returned.