Your AI is wasting compute (and you're paying for it)
Amazap gives your AI valuable shortcuts that cost less than doing it the long way: it reads a web page for about $0.01 (around 12 sats) instead of around 20 cents' worth of reading, and finds out why a build failed without reading the whole log.
Add Amazap to Claude (works on every Claude plan, including Free), then top up a few sats. Works with ChatGPT, Cursor, VS Code and more
We tested 20 web pages: 98% less for your AI to read, with nearly all the content kept. See the test
What Amazap unlocks for you
Unlock big savings on AI tokens
Valuable shortcuts that read web pages and failed builds for a fraction of what it costs your AI to read them the long way. Your AI gets just the part that matters, not the clutter around it.
- Read these five articles and tell me what they agree on. Just the text of each page, for about $0.01 (around 12 sats) a page.
- Why did my build fail? Paste the link to a failed test run and get what broke and where, for about $0.008 (around 10 sats).
- Pull the key points from this PDF. The document as clean text your AI can work with, for about $0.02 (around 24 sats).
98% less to read on 20 web pages we tested. 95% to 99.5% less on 9 real failed build logs.
Unlock things your AI can't do alone
Pay-per-call tools and data your AI can use right away, with no accounts to open and no API keys to manage. Your AI sees the price first and pays only for the calls it makes.
- A payment QR code your phone can scan. For a link or a Lightning invoice, even if your AI app can't make images. about $0.01 (around 12 sats).
- Live Lightning network data. A snapshot of a Lightning node, for about $0.01 (around 12 sats).
- Tidy page details. A page's title, summary and other key details in a neat format, for about $0.01 (around 12 sats).
Plus about 1,200 listings from other sellers that your AI can search and buy.
Unlock new revenue streams
Built a boutique model, a unique API or data nobody else has? Sell it to AI agents, paid per call, with no sign-up form between you and the sale.
- A specialist model that answers questions in your niche, one paid call at a time.
- A unique API like a tool, lookup or check only you offer.
- Your data sold one query at a time instead of in big bundles.
Every shortcut and tool shows one price before your AI uses it. That price is the whole price.
What does a web page cost your AI to read?
Example
This page is 107,522 tokens for your AI to read. Amazap's version is 1,562 tokens, 99% less.
| Model price | Raw page, per read | With Amazap |
|---|---|---|
| Budget ($0.1 per million tokens, e.g. Claude Haiku 5.5) | $0.054 (Haiku charges $0.50 above 100K tokens) | $0.010 |
| Popular ($2 per million tokens, e.g. Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 3.1 Pro) | $0.22 | $0.013 |
| Powerful ($4 per million tokens, e.g. Claude Opus 5.5) | $0.43 | $0.016 |
Prices from each provider's public price list, checked 2026-10-11. Amazap's shortcut: 12 sats (about $0.010). Counts use OpenAI's o200k tokenizer; other models count a little differently.
My AI was paying $0.22 to read this page. With Amazap, about $0.010.
Paste any public web page link. Check it, free. We fetch the page once, like a visitor who isn't signed in. Free, no sign-up.
What the wasted compute costs you today
AI models charge by the token, a small chunk of text (often part of a word). When your AI opens a web page, it doesn't just read the article. It reads all the hidden code behind the page too: menus, buttons, ads, tracking. You pay for all of it.
Here's what that looks like in real numbers from our tests, on a popular AI model that charges $2 per million tokens (the price of models like Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 3.1 Pro):
| What your AI reads | The usual way | With Amazap |
|---|---|---|
| A typical web page | 107,522 tokens: about 22 cents per read | 1,562 tokens: under half a cent to read, plus about $0.01 (around 12 sats) for the shortcut |
| A big docs page (Stripe's API page) | 597,196 tokens: about $1.19 per read | 2,384 tokens: about half a cent to read, plus about $0.01 (around 12 sats) |
| A failed build log (typical) | 79,543 tokens: about 16 cents per read | 1,855 tokens, with the cause of the failure kept |
But I don't get a bill per page. If you use a monthly plan like Claude Pro or ChatGPT Plus, you won't see these costs as a bill. If you or your company pays for AI by usage, they show up on the invoice.
When it's not worth it: if your AI already gets a clean, plain-text copy of pages, the savings are small. On the lowest-priced models, reading already costs so little that a shortcut priced about $0.01 (around 12 sats) rarely pays for itself. We'd rather tell you that up front. How we tested
How it works
- Add Amazap to your AI app. In Claude, it's one step: paste https://amazap.com/mcp as a connector and choose Approve. That works on every Claude plan, including Free. Free allows one custom connector. ChatGPT, Cursor and VS Code work too. (Amazap connects through MCP, the standard way AI apps plug in outside tools.)
- Add a little money. Ask your AI to top up. It shows a payment code you pay from any Lightning wallet, a Bitcoin payment app. A few hundred sats covers dozens of shortcuts. (A "sat" is the smallest unit of bitcoin; about $0.01 (around 12 sats).)
- Just ask. Your AI picks the right shortcut, shows you the price, and uses it. You set the limits: every purchase has a spending cap, and your AI can only spend what you added.
If something fails, you get it back. If a paid shortcut doesn't work, the price goes back to your balance as credit.
Add Amazap to Claude (works on every Claude plan, including Free), then top up a few sats.
We test our own tools, and publish the results
- Web pages: we tested 20 real pages. Page to Markdown made them 98% smaller for your AI to read and kept 94% of the main text. Read the test
- Failed builds: we tested 9 real build logs. Log Squeezer cut 95% to 99.5% of the reading and kept the cause of the failure on all 4 logs where reading the last 200 lines found none of it. Read the test
- Honest limits: some websites block automated visitors, and our page reader can't get past them. Blocked calls are credited back.
For developers: one connection, about 1,200 paid tools
The detail behind card 2. Through one MCP server, your agent can search about 1,200 pay-per-call listings (about 140 of them L402), get a quote, and buy a single call. No account with each provider and no API keys to manage.
- Quote, then buy. Every buy carries a hard cap (max_total_sats). If the real price is higher, nothing is charged.
- One balance. Top up once over Lightning; per-call and per-day limits apply.
- Failed calls credited back. Automatically when caught, or by claim within 72 hours.
- Payments: Amazap pays each seller over L402, Lightning Labs' "pay, then call" standard for APIs.
claude mcp add --transport http amazap https://amazap.com/mcp
All clients Browse the catalog Agentic payments explained
Then try: Find me a weather API on Amazap and buy one current reading for Berlin.
How selling works
Put your endpoint behind L402, paste the URL, and Amazap lists it. Anyone with a working L402 endpoint can list, from any country.
- Get paid over Lightning. We use and recommend Amboss Payments. (Amazap earns a referral credit if you sign up through this link.) Any Lightning wallet that issues invoices works.
- Put your endpoint behind L402. The starter kit has Node and Python code.
- Paste your URL. Amazap checks it and fills in the listing.
Common questions
What is Amazap, in one sentence?
Amazap gives your AI valuable shortcuts that save it money and reading: for example, it reads a web page for about $0.01 (around 12 sats) instead of around 20 cents' worth of tokens.
Do I need to be technical?
No. If you use Claude, you add Amazap once as a connector, add a little money, and then just ask your AI as usual. It picks the shortcut and shows you the price.
What does it cost?
Each shortcut has one price, shown before your AI uses it: for example about $0.01 (around 12 sats) to read a web page, or about $0.008 (around 10 sats) for a build log. Adding Amazap is free.
What are sats and Lightning?
A sat is the smallest unit of bitcoin. Lightning is a fast way to send small bitcoin payments. You pay once from a Lightning wallet app to fill your Amazap balance; your AI spends from that.
I pay a monthly plan. Does this save me anything?
You won't see a smaller bill. Your AI still reads far less: 98% less on a typical page in our test. The money savings are for people and companies who pay for AI by usage.
Can my AI overspend?
No. It can only spend what you added, every purchase has a price cap, and there are per-call and per-day limits.
What if a shortcut doesn't work?
The price goes back to your balance as credit, automatically when the failure is caught, or by claim within 72 hours.
Which AI apps work with it?
Claude, ChatGPT, Cursor, VS Code, Gemini and other apps that support MCP connectors. Setup for each app
I build agents or sell APIs. Where do I start?
Developers: install the MCP and browse about 1,200 listings. Sellers: the seller guide.
Amazap (amazap.com) is a flagship user of the Amboss API and an Amboss referral partner. For agents: llms.txt · OpenAPI · Agentic payments explained · Sell on Amazap
Turn a web page into clean text Convert a PDF to Markdown Make a QR code Check the weather Get a stock quote Look up a domain Find research papers Bitcoin and Lightning data Gift cards and eSIMs
Pay in sats, today
- API Spec Digest — 24 sats
- Feed Reader — 12 sats
- Lightning Liquidity Price Index — 24 sats
- Lightning Node Snapshot — 12 sats
Under 10 sats
- Lightning Enable: L402 Test Ping — 4 sats
- parsebit-lnd.fly.dev: Code explanation endpoint that accepts code or a code-file URL and returns a plain-English explanation via L402. — 4 sats
- api.lightningenable.com: Active weather alerts with severity scoring (0-100) via Agent Commerce Store L402 proxy. Listed price: 2 sats/request. — 5 sats
- api.lightningenable.com: Current conditions from nearest observation station (F and C) via Agent Commerce Store L402 proxy. Listed price: 2 sats/request. — 5 sats
Bitcoin data
- mcp.loopxxi.com: loop-mcp REST: btc_price (10 sats) — 13 sats
- api.lightningenable.com: Agent Commerce Store YouTube transcript API endpoint. GET returns HTTP 402 with L402 macaroon and Lightning invoice. — 8 sats
- faucet.mutinynet.com: Mutinynet Faucet L402 Auth — 60 sats
- Raikiri: Bitcoin transaction lookup, decoded (own full node) — 13 sats
Catalog
- Log Squeezer — 10 sats
- Package Facts — 20 sats
- Page Metadata — 12 sats
- Page to Markdown — 12 sats
- PDF to Markdown — 24 sats
- QR Code — 12 sats
- Repo Digest — 24 sats
- L402 Agent Mail — 60 sats
- L402 Agent Toolkit — 8 sats
- L402 LNURL Resolve — 13 sats
- L402 Mempool Fee Oracle — 13 sats
- L402 Nostr Event Verify — 13 sats
Amazap Basics
- API Spec Digest — 24 sats
- Feed Reader — 12 sats
- Lightning Liquidity Price Index — 24 sats
- Lightning Node Snapshot — 12 sats
- Log Squeezer — 10 sats
- Package Facts — 20 sats
- Page Metadata — 12 sats
- Page to Markdown — 12 sats
Popular right now
- QR Code — 12 sats
- PDF to Markdown — 24 sats
- Repo Digest — 24 sats
- Weather Intel: Forecast — 6 sats
- YouTube Transcript: Extract — 8 sats
- Bitrefill — esims search — 6 sats
- Bitrefill — gift-cards search — 6 sats
- Bitrefill — topups search — 6 sats
New and tested
- lightning Rebalance Cost Index (realised ppm, 7d/30d) — 55 sats
- api.lightningenable.com: Current conditions from nearest observation station (F and C) via Agent Commerce Store L402 proxy. Listed price: 2 sats/request. — 5 sats
- Raikiri radiation API: daily mean/min/max background radiation (Japan, since 2022) — 13 sats
- Raikiri radiation API: hourly background radiation history (Japan) — 13 sats
- Research Suite: Multi-Source Search — 13 sats
- Raikiri: L402 mystery-shopper results (detail) — 23 sats
- TOON anonymous fetch (l402-door) — 8 sats
- L402 LLM API (pyfile-toolkit) — 20 sats