The trusted context layer for developers and AI agents.
Search, scrape, and extract any URL into clean markdown or typed JSON, for the software you build and the agents you deploy. Every structured field carries a confidence score and a source trail, so you act on facts, not guesses.
Free tier · 1,000 credits/mo · no card required
/v1/scrape · Any URL to clean, LLM-ready markdown.
Pick an operation, enter a url, and press Run. Structured results show a confidence score and a source trail.
Live against the real API. Structured results ship a confidence score and a source trail.
Built for agents. They mint their own key.
An agent should not wait on a human to fill out a signup form. Yours calls POST /v1/keys/provision, reads a SKILL.md, and starts scraping inside its own loop. Headless key minting, no dashboard click-through.
- POST /v1/keys/provisionThe agent mints its own free-tier key in one call. No human, no dashboard, no signup form to click through first.
- GET /agent-onboardingThe agent fetches a SKILL.md and learns the whole API from one file. No docs to read, no human to ask.
- mcp.superscraper.devPoint Claude, Cursor, or any MCP-aware agent at the hosted server with one config block. It has live web access, and you host nothing. Live at launch.
# 1. The agent mints its own key, no human signup
curl -X POST https://superscraper-production-1381.up.railway.app/v1/keys/provision \
-H "Content-Type: application/json" \
-d '{ "label": "research-agent" }'
# → { "apiKey": "ss_live_...", "tier": "free" }
# 2. Use it immediately, fetch + trust the result
curl -X POST https://superscraper-production-1381.up.railway.app/v1/scrape \
-H "Authorization: Bearer ss_live_..." \
-d '{ "url": "https://example.com", "formats": ["markdown"] }'
# → { "markdown": "...", "_confidence": 0.97, "_extraction_method": "json-ld" }Stop making your agent guess
A black-box scraper hands back data and walks away. Your agent cannot tell a solid read from a lucky one. We attach a confidence score and a source trail to structured results, so the model decides whether to trust a field, retry, or escalate. When a site publishes structured data we read it directly, so it is exact, not estimated.
A structured result reports how sure it is (0 to 1). The agent branches on the number: accept it, retry, or escalate to an LLM pass.
JSON-LD, Open Graph, regex, or an LLM. You see how each field was read, so a black box becomes a decision you can audit.
The fetch tier that won and how the result was obtained, attached to the payload. A source trail, not a guess.
What you get that the others don't
Clean markdown is table stakes now. The difference is whether a result tells your agent how it was read and how sure it is. Here is the honest line-up.
| Capability | SuperScraper | Firecrawl | Jina Reader | Proxy APIs |
|---|---|---|---|---|
| Clean LLM-ready markdown | Yes | Yes | Yes | Raw HTML |
| Structured extract | Free JSON-LD | Paid | No | No |
| Confidence score per result | Yes | No | No | No |
| Source trail (provenance) | Yes | No | No | No |
| Agent self-provisions its key | Yes | No | No | No |
| Hosted MCP server | Yes | Yes | No | No |
| Failed scrape billed | Never | Varies | Not stated | Yes |
As of 2026-08-02. Proxy APIs = ScrapingBee, ScraperAPI, and similar. Rows reflect each product's standard scrape response, not add-on services. Firecrawl and Jina are trademarks of their owners; comparison is for orientation, not endorsement.
Eight endpoints, one key
Scrape, search, crawl, map, extract, batch, parse, screenshot. Every way to pull the live web, behind one /v1 key. No separate services to wire together.
One URL becomes clean markdown, raw HTML, links, a screenshot, or structured JSON. Cookies, headers, waitFor, and browser actions included.
No URL? Describe it. Search the web and scrape the results in one call, straight into LLM-ready content.
All the content on a site, recursive. Include and exclude paths, depth limits, async jobs with completion webhooks.
All the links on a domain, no content. Sitemap plus crawl graph in one call.
Pass a schema, get typed JSON back. One URL or a bounded multi-URL fan-out.
Up to 100 URLs per request, sync or async, with per-tier concurrency.
A PDF or DOCX (upload or URL) becomes markdown an agent can read. SSRF-guarded.
Full-page or viewport captures, mobile emulation, delivered to storage or base64.
Output a model can read
Raw HTML is noise to a model: nav, ads, scripts, half-rendered JS. We return any URL as markdown or typed JSON with the boilerplate stripped. Your agent reasons over the content, and burns far fewer tokens doing it.
- Markdown, HTML, links, JSON: pick the formats you need per request
- Free JSON-LD and Open Graph extraction: no LLM cost when a site already ships structured data
- Content shaping: main content only, include and exclude tags, max-age cache
- Anti-bot that gets out of your way: plain fetch, then stealth browser, then proxy rotation, only as far as a site needs
curl -X POST https://superscraper-production-1381.up.railway.app/v1/scrape \
-H "Authorization: Bearer ss_live_..." \
-H "Content-Type: application/json" \
-d '{
"url": "https://example.com/pricing",
"formats": ["markdown", "json"]
}'Drop it into the loop you already have
Claude, Cursor, or your own agent. Point it at the hosted MCP and the model has live web access, with no scraping infrastructure to run.
Hosted MCP
A hosted MCP server for Claude, Cursor, and any MCP-aware agent. Add the server, and a model has live web access. You host nothing.
Plain HTTP, no SDK needed
Every endpoint is REST with bearer auth, so curl or your language’s HTTP client is enough. On a Firecrawl-shaped client already? Point it at our base URL and it ports directly. First-party SDKs are in progress and not published yet.
SKILL.md and llms.txt
A SKILL.md the agent reads to learn the API, plus llms.txt generation for any site so a model can ground itself fast.
Simple, predictable credits
One page is one credit. JSON-LD and Open Graph extraction is free. A scrape that returns no data costs nothing.
- 1 credit= 1 page scraped
- $0free JSON-LD / Open Graph extraction
- $0failed scrapes, never billed
- 1,000/mofree tier, no card
Ship your first scrape in five minutes
Create a key, or let your agent mint its own. Paste the cURL, get structured JSON back with a confidence score. No sales call, no credit card.