Your agent can read a webpage — until the page needs a browser, the link hides behind six redirects, or the data lives somewhere with no API. fetcher slips through the wall and brings it back. One HTTP call, paid per use, no keys.
Works with any tool-using agent — Claude · ChatGPT · Gemini · Grok · Perplexity · MCP
# your agent hits a JavaScript-only page it can't read $ curl "https://api.fetcher.cloud/render?url=app.example.com&screenshot=true" # first call → the ghost asks to be paid HTTP/1.1 402 Payment Required { "accepts": [ Base · USDC · $0.02, Solana · USDC · $0.02 ] } # agent pays a fraction of a cent, retries, gets the rendered page HTTP/1.1 200 OK { "title": "...", "html": "<fully rendered DOM>", "screenshot_png_base64": "..." }
Large language models are brilliant at reasoning and blind at retrieval. The moment real-world data needs a browser, a payment, or a login, the agent stalls. fetcher is the specialist it calls.
Each is a single GET request that returns clean JSON. Start free every day; pay only past the free tier. These four are the flagships — the full lineup is in the pricing table below.
A real Chromium browser loads the page, waits for it to settle, and returns the fully-rendered HTML, visible text, and an optional screenshot. The one thing plain fetch() can never do.
Readability as a service. Strips nav, ads, and boilerplate and returns the main article as Markdown and text, with title, author, headings, links, and reading time — built for RAG ingestion.
Unwraps shortened, tracking, and wrapped URLs. Returns the final destination, every hop with status codes, and the landing page's metadata — so an agent knows where a link goes before it follows it.
Real-time availability across TLDs via RDAP. The enriched tier adds DNS records, mail and email-auth signals, domain age, and Wayback Machine history for vetting a domain before you buy it.
No dashboard, no key rotation, no invoices. The agent and the endpoint settle it between themselves using x402 — the open standard that turns HTTP 402 into a real payment.
The agent makes a normal GET request. The first unpaid call returns 402 Payment Required with machine-readable terms.
The agent's wallet pays in USDC on Base or Solana — whichever chain it holds — and automatically retries. No human approval.
The ghost returns exactly what was asked for: rendered HTML, Markdown, a redirect trace, or domain data. Ready to use.
Every endpoint advertises its price, schema, and free tier at GET /. Agents discover what fetcher does without a human reading a page.
Callable directly inside agent sessions built on the Model Context Protocol, or over plain HTTP from any framework — LangChain, custom loops, whatever you run.
Priced tools are marked discoverable for x402 indexes, and an llms.txt tells AI crawlers exactly what lives here and how to use it.
Every tool has a free daily allowance. Past it, you're billed per call in USDC — often less than a tenth of a cent.
| Endpoint | What you get | Free / day | Per call |
|---|---|---|---|
| Browser | |||
| /render | Rendered HTML, text, screenshot | 10 | $0.02 |
| /screenshot | Base64 PNG/JPEG screenshot | 10 | $0.02 |
| Content | |||
| /extract | Clean article Markdown + metadata | 30 | $0.005 |
| /unfurl | Link preview card (OG/Twitter) | 50 | $0.003 |
| /feed | RSS/Atom feed → JSON items | 40 | $0.003 |
| /oembed | Embed data (YouTube, Vimeo, X…) | 50 | $0.002 |
| /pdf-text | Text + metadata from a PDF URL | 20 | $0.01 |
| /diff | Compare two pages / vs Wayback | 25 | $0.008 |
| Web intel | |||
| /resolve | Redirect chain + metadata | 50 | $0.003 |
| /headers | HTTP + security-header analysis | 50 | $0.002 |
| /robots | robots.txt rules + sitemaps | 50 | $0.002 |
| /favicon | Best site icon / logo | 50 | $0.002 |
| Domain & DNS | |||
| /domain | RDAP availability across TLDs | 50 | $0.004 |
| /domain/enriched | + DNS, age, Wayback history | 25 | $0.01 |
| /whois | Deep registration record | 40 | $0.005 |
| /dns | Full DNS record dump | 50 | $0.003 |
| /ssl | TLS cert issuer, expiry, SANs | 40 | $0.004 |
| Utility | |||
| /qr | Generate a QR code (SVG) | 50 | $0.002 |
Most agents can only fetch raw HTML. They can't run JavaScript to see single-page apps, can't safely follow shortened or tracking links to their real destination, can't strip a page down to clean article text, and can't query live domain-registration data. fetcher does each of these behind one paid HTTP call, and returns structured JSON an agent can use directly.
Any agent that can make an HTTP request and hold a wallet. That includes agents built on Claude, ChatGPT, Gemini, Grok, Llama, Mistral, and Perplexity, as well as custom agents using frameworks like LangChain or the Model Context Protocol. fetcher is model-agnostic — it's just an endpoint.
No. Payment is the authentication. The first call returns HTTP 402 with terms; your agent pays in USDC on Base or Solana and retries. Every tool also has a free daily tier for testing and low-volume use, gated per caller.
Autonomous agents can't sign up for SaaS plans or enter credit cards. Sub-cent, per-call settlement lets an agent pay for exactly what it uses, the instant it uses it — which is the only billing model that works when the customer is software.
Point it at an endpoint and watch it fetch what it couldn't reach a minute ago.