Creating a proxy sub-account, changing an IP whitelist, resetting a capture key and buying static IPs all gate behind that two-step confirm — the right default when the account being changed is the one paying for every other call. Beyond that, the surface is organised around picking the right granularity: search for a lookup, research when you want passages from independent sources, extract for a URL you already hold, crawl for a bounded set, site copy when the answer is a whole knowledge base on disk. The per-platform scrapers run the same engine pinned to one platform with a typed enum, so choosing between them is mostly about typing safety.
A web-data server spanning search, extraction, crawling and whole-site mirroring, with per-platform structured scrapers, residential proxies, browser automation, and account and billing management exposed as tools. It is hosted-first — point a client at one URL and install nothing — with the same package runnable locally when you need the tools the hosted endpoint cannot offer.
- Search with the reading folded in: novada_search returns reranked titles, URLs and snippets with domain, date and geo filters and an option to auto-extract the top result; novada_research runs parallel searches across Google, Bing and DuckDuckGo, dedups, extracts full content and returns numbered cited source sections for you to compose from.
- Extraction that escalates by itself: novada_extract moves from static fetch to JS render to a real browser against Cloudflare, DataDome and Kasada, takes a list of URLs in parallel, and returns markdown or structured JSON under a character ceiling.
- Site-shaped jobs at three scales: novada_map returns what URLs exist from the sitemap or a crawl fallback, novada_crawl walks a bounded set of pages extracting content inline with path filters, and novada_site_copy writes an entire docs site to disk as one markdown file per page and returns a compact manifest rather than the bodies.
- Structured records instead of raw HTML: novada_scrape covers sixteen platforms, with a dedicated closed-enum tool per platform — novada_scrape_amazon, _google, _bing, _duckduckgo, _yandex, _youtube, _instagram, _facebook, _tiktok, _x, _walmart, _shein, _linkedin, _github and _perplexity — so an agent cannot invent an operation the platform does not have.
- Interaction and network control: novada_browser gives full browser control and chains actions in one call, novada_browser_flow offers a simplified five-action version, and novada_proxy returns a residential or mobile proxy URL, shell exports or a curl flag with city, country and a sticky session.
- Checking and watching: novada_verify runs searches from supporting, skeptical and fact-check angles and reports whether a claim is supported, unsupported, contested or undecidable; novada_monitor hashes a page and reports field-level diffs against the previous check; novada_ai_monitor searches public pages on AI-company domains for brand mentions and sentiment.
- Account and infrastructure as first-class tools: novada_setup validates the key and shows the balance at no credit cost, novada_account composes wallet balance, plan quotas and entitlements by section, and the proxy sub-account, IP whitelist, capture-key and static-IP tools manage the infrastructure behind it.
- novada_discover lists every tool with its name, description, category and status — the call to make on your own connection, because the hosted and self-hosted surfaces differ.
A Novada account — free to start, no card, with credits capped at a thousand calls a month. Nothing to install: point your client at the hosted URL, whose token parameter carries your key, which is why that URL should be treated as a password and never configured as a shared or organisation-level connector. One thing decides whether hosted is enough: the stateless endpoint leaves out eight tools that need a persistent session, disk, or write-gated account access — among them the site copy, the simplified browser flow, the claim verifier and the account-management tools. Run it locally with NOVADA_API_KEY set if you need those.
One command plus a key — npx -y novada-mcp, then supply credentials