Sesame: Web Unblocker API & MCP for AI Agents | FetchLayer

Sesame · Web Unblocker

Say the word.
Any page opens.

Give FetchLayer Sesame a public URL and get the page back: title, text and links. That includes the pages that answer a script with a 403.

Press Open Sesame, or paste a URL of your own.

Try another page

30 free requests, no credit card. From 1 credit. Nothing for a page we can't open.

A URL goes in. The page comes out.

Sesame is one API call. This is all of it.

  1. 1

    You send a URL

    POST /web-unblocker/fetch

    { "url": "books.toscrape.com/…/a-light-in-the-attic" }

    From your code over REST, or from your AI agent as an MCP tool.

  2. 2

    Sesame opens the page

    Any page1 credit

    Couldn't be opened0 credits

    Any public page, including the ones that turn scripts away. You don't choose how hard it tries.

  3. 3

    You get the page back

    A Light in the Attic

    Price
    £51.77
    Availability
    In stock (22 available)

    3 links1 credit370 tokens

    In the form you ask for: htmltextmarkdownreadablelinksmetadatascreenshot

curl gets a 403. Sesame gets the page.

Two requests to npmjs.com/package/zod, minutes apart on 10 November 2026. Nothing below is mocked up.

curl npmjs.com/package/zod403

Just a moment...

Enable JavaScript and cookies to continue

5,447 bytes, none of them the page.

POST /web-unblocker/fetch200

zod - npm

TypeScript-first schema declaration and validation library with static type inference. Latest version: 4.6.7, last published: a month ago.

What is Zod?

Zod is a TypeScript-first validation library. Define a schema and parse some data with it. You'll get back a strongly typed, validated result.

1,331 tokens of content and 13 links, for 1 credit.

What comes back.

One page, in the form you ask for

The product page from the demo above, four ways. Markdown, plain text and a screenshot are there too.

# A Light in the Attic

£51.77

In stock (22 available)

## Product Description

It's hard to imagine a world without A Light in the Attic. This now-classic collection of poetry and drawings from Shel Silverstein celebrates its 20th anniversary with this special edition.

## Product Information

| Price (incl. tax) | £51.77 |
| Availability | In stock (22 available) |

370 tokens

Sized for a model

That page is 1,879 tokens as HTML and 370 as readable: 80% less for your model to read, with the price and stock still in it.

A block page is never your answer

Name the text that must be on the page. If a "please verify" page comes back instead, the request fails honestly and costs nothing.

Your agent can use it in one line

claude mcp add --transport http fetchlayer https://mcp.fetchlayer.dev

Works in Claude, ChatGPT, Cursor and any other MCP client. Same key, same credits as the REST API.

Not opened, not billed

You never pay to find out a page is out of reach. Set maxEffort and you also cap what any one page may cost.

Three things teams do with it.

Price monitoring

The price on the page, every time you ask.

Product and pricing pages are the ones that most often refuse a script. Name the text that has to be there and you get the real page or an honest failure, never a "please verify" page saved as a price.

expectText
"Add to cart"
country
"de"
maxEffort
"standard"

The page as a visitor in Germany sees it, or an unbilled failure.

AI agents and RAG

A page that fits in a context window.

Ask for readable and the page arrives as its main content in compact Markdown: about 8,000 tokens where the HTML ran to 100,000. Pass a question and only the parts about it come back.

output
"readable"
query
"refund policy"
maxTokens
2000

The sections that answer the question, with a cursor for the rest.

Research

A list of URLs in, a table out.

Sweep ten thousand addresses with a low ceiling, then re-run only the ones that needed more. Every answer says what it cost, so the bill for a study is something you add up, not estimate.

output
"metadata"
maxAge
600
maxEffort
"basic"

Title, description, dates and structured data per page.

One POST.

Send a URL. Get the page, and what it cost, in the same response. Or skip the code and give your agent the tool.

curl -X POST https://api.fetchlayer.dev/web-unblocker/fetch \
  -H "Authorization: Bearer $FETCHLAYER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html", "output": "readable"}'

What comes back

{
  "title": "A Light in the Attic | Books to Scrape - Sandbox",
  "status": 200,
  "contentFormat": "readable",
  "content": "# A Light in the Attic\n\n£51.77\\n\\nIn stock (22 available)…",
  "totalTokens": 370,
  "shownTokens": 370,
  "pagesFetched": 1,
  "usage": { "billableUnits": 1, "totalMs": 1410 }
}

Used by developers at

Individual developers or teams — not official partnerships

From 1 credit.

A page you got costs credits. A page you didn't get costs nothing. You never pick a tier: the response tells you what the page cost, in the same credits as every other FetchLayer API.

Three real pages from the demo above. One of them refuses a plain request. All three cost 1 credit.

Product pagebooks.toscrape.com

200Opens to a plain request

1 credit

A real shopikea.com

200Opens to a plain request

1 credit

A hard pagenpmjs.com

403Refuses a plain request

1 credit

1credit

Any page

Ordinary pages and the ones that turn scripts away, at the same price.

$1.99 per 1,000 pages

0credits

Not opened

A page we could not open is never billed.

$0.00, always

5 for the very hardest pages: rare, and only when it was actually needed.

Dollar figures are at the pay-as-you-go rate of $1.99 per 1,000 credits, and a plan lowers them. Set maxEffort to cap what any one request may cost. See the plans

Questions.

What does FetchLayer's web unblocker API return, and what does it cost?

FetchLayer Sesame (Sesame · Web Unblocker) is our web unblocker API: send any public URL and get the page back as JSON, as readable content sized for a model, Markdown, visible text, HTML, its links, its metadata or a screenshot, including pages that turn an ordinary request away. From 1 credit, and most requests cost exactly that: a simple page, one that fills in its content after it loads or one that turns ordinary requests away is 1 credit. 5 credits applies only to the very hardest pages, rare in practice, and only when it was actually needed. You are charged at the level the page needed, and a request that returns no page is not billed. Set maxEffort to advanced and a request never costs more than 1 credit; a page that needs more comes back unsuccessful and is not charged. Credits cost $1.99 per 1,000 pay-as-you-go, or $25 a month buys 15,000, with 30 free to start. Both routes are also tools on our MCP server at https://mcp.fetchlayer.dev, for Claude, ChatGPT, Cursor and other MCP clients.

What is Sesame?

FetchLayer Sesame is our web unblocker: the same API, key and credits as before, under its own name. Existing routes, MCP tool names and integrations are unchanged.

What is a web unblocker?

One HTTP endpoint that fetches any public URL and hands back the page — as HTML, visible text, Markdown or an image. It gets the pages an ordinary fetch cannot: the ones that come back empty, and the ones that turn a plain request away. You send a URL with a Bearer token; you get the page, and a receipt saying what it cost.

What does a page cost?

From 1 credit, and most requests cost exactly that: a simple page, one that fills in its content after it loads or one that turns ordinary requests away is 1 credit. 5 credits applies only to the very hardest pages, rare in practice, and only when it was actually needed. You are charged at the level the page needed. The exact figure is on every response as pagesFetched, so you never have to guess. A request that comes back with no page is not billed at all.

Can I cap what a request is allowed to cost?

Yes, and that is what maxEffort is for: a ceiling on how hard a single request may work for the page before it gives up. Sweeping ten thousand URLs where most are plain HTML? Set the lowest ceiling and re-run only the failures without it. A capped request says so in notes, so it is never mistaken for a broken page. Set maxEffort to advanced and a request never costs more than 1 credit; a page that needs more comes back unsuccessful and is not charged.

How long can one request take?

There is a hard 120-second ceiling on a single request. Work that cannot finish inside what is left of it is abandoned before it starts rather than cut off halfway — so you are never billed for effort that produced nothing. A 404 comes straight back: the page is not there, and working harder cannot change that.

What happens when nothing works?

A 503, and it is not billed — a request that produced no page costs nothing, so a hard target is never something you pay to discover. The body carries the reason it gave up, and lowering maxEffort and re-running is how you narrow down which pages in a batch are the expensive ones.

Are the hard pages as fast as the easy ones?

No, and it is worth planning for. A page that gives up its content immediately comes back in well under a second. The most stubborn ones take considerably longer, and in a large burst some will run out the 120-second ceiling and come back unread rather than wrong. Expect a small slow tail rather than a uniform latency, and retry that tail rather than sizing every timeout for it.

How do I know the charge is right?

You read it rather than trust it. Every response that returns a page carries a usage block: what the request cost, the effort level that served the page (which is what sets the price), bytes transferred and the wall-clock total. A request that cost more than you expected tells you which level it needed, on the response, at the time.

How do I avoid saving a block page by mistake?

A blocked request that answers 200 with a "please verify" page is the failure that quietly poisons a dataset. Two parameters stop it: expectText names text that must appear, and waitForSelector names an element that must be present. If what you asked for never appears, the request is reported unsuccessful rather than answered with the wrong content.

What will it refuse to fetch?

Public http and https addresses only, checked before a request is made rather than after. Reserved address space is refused — loopback, private ranges, link-local and cloud metadata — along with every spelling of it: integer forms such as http://2852039166/, IPv4-mapped, NAT64 and 6to4. Ports are allowlisted to 80, 443, 8080 and 8443. Credentials before the hostname are rejected. Every redirect hop is re-validated against the same rules as the first, so a public address cannot redirect into a private one.

What is not implemented?

Two things, said plainly rather than left to be discovered. Nothing drives the page for you — no click, scroll or type; waitForSelector, waitMs and waitUntil are the whole vocabulary for waiting. And there is no per-target-host rate limiting: if you need to be gentle with one site, pace it yourself.

Can my AI agent use it?

Yes. Both routes are MCP tools on the same FetchLayer MCP server, named web_unblocker_fetch and web_unblocker_screenshot. Ask for output: "readable" and the page arrives as its main content in compact markdown, about 8,000 tokens or fewer where the HTML runs to 100,000 or more, with totalTokens and shownTokens saying how much came back. Tool calls use the same key and are billed exactly like REST.

Open Sesame: start free.

30 free requests, no credit card. FetchLayer Sesame runs on the same key and credits as every other FetchLayer API.