I built HolidayIQ, a free holiday decision API on Cloudflare Workers. It wraps Nager.Date (keyless, free) so apps can answer "should I send this email/promo today?" for any country. One practical use: checking whether a country has a public holiday before sending a marketing email. No point promoting a Christmas sale in a country where Dec 25 isn't a holiday. Stack: * Cloudflare Workers * Nager.Date v3 (https://date.nager.at/api/v3) * RapidAPI listing (free $0 BASIC plan): https://rapidapi.com/brainsai1109/api/holidayiq Live API: https://holiday-iq-api.brains-ai1109.workers.dev Landing page: https://holidayiq-holidayiq-is-holiday-decision-layer-t4w2kn3kch.openbot.site BASIC is $0, no credit card required. Paid plans later, maybe. For now it's just shipping — feedback welcome.

From teenage hackathons to RapidAPI’s $1 billion valuation, a CEO departure and Gamut: the career and practical software ideas of Iddo Gino.

When I first started connecting external tools to AI agents, I assumed an API marketplace had already solved the problem.It solved only half of it. ## The API Marketplace Model Take RapidAPI as a familiar example. The workflow is roughly: Developer → Search for an API → Read docs → Choose a plan → Get credentials → Write integration code → Call API This works well when a developer already knows what capability the application needs. Need weather data? Find a weather API. Need stock prices? Find a market-data API. The marketplace reduces discovery, billing, and integration friction, but the developer still owns most of the orchestration. ## What Changes With AI Agents? Now imagine the client is an agent. The user says: > Analyze ETH performance over the last 30 days > and explain the largest price movements. The agent may need to: 1. discover a suitable market-data service; 2. inspect its inputs and pricing; 3. decide whether it fits the task; 4. invoke it; 5. potentially call another service; 6. combine the results. That is a different abstraction. Instead of: Developer → API Marketplace → API the architecture starts looking like: User Intent ↓ AI Agent ↓ Service Discovery ↓ Select + Invoke ↓ External Service ↓ Result ## Agent Service Marketplace Anvita Flow is one example of this emerging model. Rather than treating the marketplace mainly as a catalog for developers, the service layer is designed to be consumed by agents: discover capabilities, use external services, coordinate specialized agents, and handle spending controls within the workflow. The important difference isn't that APIs disappear. They don't. The difference is who handles the integration logic. ## My Takeaway API marketplaces are still useful infrastructure. But once agents begin choosing tools dynamically, the missing layer isn't another API directory. It's a runtime service layer that can answer: “Given this goal, which capability should I use next?” That may be the more interesting marketplace problem for agent developers. How are you handling service discovery in your agents today: hard-coded tools, MCP, an API marketplace, or something else?

Explore your APIs

RapidAPI for Mac est officiellement abandonné, comme Nokia (oui, le même que les 3310) l’a discrètement annoncé sur le site officiel du projet. Ce client HTTP né en 2013 sous le nom de Paw était pourtant encore très populaire dans le monde du développement web. Cette interface graphique native pour macOS améliorait l’utilisation d’un outil comme cURL pour interroger des API et tester différentes configurations. L’app sert au quotidien au sein de l’équipe technique de MacGeneration… qui va devoir trouver une alternative à terme. RapidAPI en action, ici pour tester l’API de l’iTunes Store. Image MacGeneration. Pour le moment, RapidAPI fonctionne encore normalement, y compris sous macOS 27 qui vient de sortir. Néanmoins, ses propriétaires indiquent déjà que Golden Gate ne fait pas partie des versions officiellement prises en charge, ce qui signifie qu’il ne faut pas s’attendre à des correctifs ou améliorations si c’était nécessaire. Pour continuer à recevoir de l’aide de la part de Nokia, il faut rester sur une version gérée, c’est-à-dire macOS 26 au maximum. Au-delà de l’assistance, cela signifie surtout que le développement de l’app est déjà plus ou moins interrompu, ce qui la condamne forcément sur la durée. Même si RapidAPI reste fonctionnel aujourd’hui et devrait encore s’ouvrir normalement dans les mois qui viennent, les mises à jour de macOS finiront par ajouter des bugs qui rendront l’app de plus en plus inutilisable. Si ses utilisateurs n’ont pas été informés directement de la décision, ils devront toutefois se faire une raison et commencer à chercher une alternative. Image MacGeneration, photo de base Julia Kadel sur Unsplash. L’option la plus connue est sans doute Postman, même si ce n’est plus une app native aussi légère. En faisant quelques recherches rapides, mais sans avoir testé (je ne suis pas développeur, après tout), j’ai repéré Bruno et Insomnia qui présentent l’avantage d’être open source. Si vous avez de meilleurs conseils à proposer, n’hésitez pas à recommander vos alternatives dans les commentaires.

Explore your APIs

A useful trend-research workflow rarely ends with a hashtag count. You usually need to move from a hashtag to recent public videos, then to the creators behind those videos—without silently accepting duplicate pages or anti-bot responses as real data. This tutorial builds that workflow with four read-only endpoints: * `/v1/hashtag/info` * `/v1/hashtag/videos` * `/v1/users/info` * `/v1/videos` The examples use the TikTok Public Metadata API on RapidAPI. It has a free BASIC plan with 150 requests per month, which is enough to run this tutorial and test the response contract. > This is an independent public-web metadata API, not TikTok's official API. > Use public data lawfully and respect privacy, intellectual-property rights, > TikTok's applicable terms, and RapidAPI's terms. ## 1. Create a small API client Store your RapidAPI key in an environment variable. Never commit it. export RAPIDAPI_KEY="your-key" const BASE_URL = "https://tiktok-public-metadata-api.p.rapidapi.com"; const headers = { "X-RapidAPI-Key": process.env.RAPIDAPI_KEY, "X-RapidAPI-Host": "tiktok-public-metadata-api.p.rapidapi.com" }; const sleep = (milliseconds) => new Promise((resolve) => setTimeout(resolve, milliseconds)); async function apiGet(path, parameters, attempt = 0) { const url = new URL(path, BASE_URL); for (const [name, value] of Object.entries(parameters)) { if (value !== undefined && value !== null) { url.searchParams.set(name, String(value)); } } const response = await fetch(url, { headers }); if ((response.status === 429 || response.status === 503) && attempt < 4) { const retryAfter = Number(response.headers.get("retry-after")); const delay = Number.isFinite(retryAfter) && retryAfter >= 0 ? retryAfter * 1000 : 500 * 2 ** attempt + Math.random() * 250; await sleep(delay); return apiGet(path, parameters, attempt + 1); } const body = await response.json(); if (!response.ok) { throw new Error( `${response.status} ${body?.data?.error?.code ?? "API_ERROR"} ` + `(request_id=${body?.request_id ?? "unknown"})` ); } return body; } TikTok can return HTTP 200 challenge or anti-bot payloads on public web routes. The API validates the native status, payload shape, and requested resource identity; an untrustworthy response becomes a retryable structured `503` instead of a fake empty success. ## 2. Read exact hashtag metadata The leading `#` is optional: const hashtag = await apiGet("/v1/hashtag/info", { tag: "music", raw: false }); console.log({ requestId: hashtag.request_id, title: hashtag.data?.title, videoCount: hashtag.data?.statistics?.video_count, viewCount: hashtag.data?.statistics?.view_count }); Keep `request_id` in your logs. It lets support trace one request without logging your API key or a full native payload. ## 3. Fetch videos and preserve the returned cursor Start with `cursor=0`. Do not increment, decode, edit, or reconstruct the cursor returned by a list endpoint. async function getHashtagPage(tag, cursor = 0, count = 10) { const page = await apiGet("/v1/hashtag/videos", { tag, cursor, count, raw: false }); if (page.count !== page.data.length) { throw new Error("Unexpected page contract"); } const ids = page.data.map((video) => video.id); if (new Set(ids).size !== ids.length) { throw new Error("Duplicate video IDs in one page"); } return page; } const firstPage = await getHashtagPage("music", 0, 10); console.log({ count: firstPage.count, hasMore: firstPage.has_more, nextCursor: firstPage.cursor }); Continue only when `has_more === true`, copying the returned cursor unchanged: const secondPage = firstPage.has_more ? await getHashtagPage("music", firstPage.cursor, 10) : null; The cursor is signed and bound to the original hashtag and page state. It can also preserve unconsumed items when the API merged more than one native TikTok window to fill a requested page. ## 4. Turn video results into creator rows Normalize creator usernames from the returned videos, then fetch each public profile once. Keep concurrency low so a research script remains friendly to both the API and the upstream service. const usernames = [ ...new Set( firstPage.data .map((video) => video.author?.unique_id) .filter(Boolean) ) ]; const creators = []; for (const username of usernames.slice(0, 5)) { const profile = await apiGet("/v1/users/info", { username, raw: false }); creators.push({ username: profile.data?.unique_id, nickname: profile.data?.nickname, followers: profile.data?.statistics?.follower_count ?? null, videos: profile.data?.statistics?.video_count ?? null }); } console.table(creators); Unknown values remain `null`. Do not convert them to `false` or `0` unless your own application explicitly defines that meaning. ## 5. Add one creator's public video feed Use the same cursor rule for creator videos: if (creators[0]?.username) { const creatorVideos = await apiGet("/v1/videos", { username: creators[0].username, cursor: 0, count: 5, raw: false }); console.log( creatorVideos.data.map((video) => ({ id: video.id, description: video.description, views: video.statistics?.play_count ?? null })) ); } At this point the workflow can answer practical questions: * Which public videos are associated with a hashtag? * Which creators appear in those results? * What public audience and posting signals are available for those creators? * What should be collected on the next page without duplicating items? ## Production notes * Retry `429` and temporary `503` responses with backoff and jitter. * Do not retry `400`, `401`, or `404` unchanged. * Cache normalized results for your own product when appropriate. * Request `raw=true` only when you need a native TikTok field that is not normalized; native payloads are larger and can change. * Preserve `request_id` for support, but never log the RapidAPI key. The production monitor currently runs 37 customer-path probes through the RapidAPI gateway, covering all 16 data endpoints, multiple creators and hashtags, pagination, raw responses, known regressions, and empty final pages. ## Try the workflow The free BASIC plan includes 150 requests per month: https://rapidapi.com/pomiandaitumm/api/tiktok-public-metadata-api?utm_source=devto&utm_medium=content&utm_campaign=tiktok_api_hashtag_research_202608 Full documentation, live status, limitations, and OpenAPI: https://tiktok-public-metadata-api.pomiandaitumm.chatgpt.site If you build creator analytics or social-listening tools, I would especially like feedback on which normalized hashtag and creator fields save you the most integration work.

Unofficial Tapology MMA API: Access fighter profiles, upcoming event schedules, and fight history. • Comprehensive fighter profiles with details and stats • Upcoming Event schedules per subregion • Up-to-date upcoming, cancelled and past fight results • Coverage of both amateur and professional fighters worldwide Powered by Tapology, the world's largest fight database, for real-time accuracy.

Unofficial Tapology MMA API: Access fighter profiles, upcoming event schedules, and fight history. • Comprehensive fighter profiles with details and stats • Upcoming Event schedules per subregion • Up-to-date upcoming, cancelled and past fight results • Coverage of both amateur and professional fighters worldwide Powered by Tapology, the world's largest fight database, for real-time accuracy.

Scraping Twitter/X without direct API access can be accomplished by leveraging third-party services like RapidAPI, which provides a streamlined way to interact with Twitter data. In this tutorial, we'...

Redefining how developers work with APIs. One Git-native, composable workspace. Offline-first. Markdown-powered. Developer-loved.