WebLens

Premium web intelligence API for AI agents, with x402 micropayments. No accounts, no API keys — pay per request in USDC on Base.

🌐 Live API: https://api.weblens.dev 📖 Interactive docs: https://api.weblens.dev/docs 🤖 Agent guide: https://api.weblens.dev/llms.txt 🔎 Marketplace: listed on x402scan

Try it free, right now

No wallet, no signup — these are rate limited to 10 requests/hour per IP:

curl https://api.weblens.dev/r/https://example.com     # any page as markdown
curl https://api.weblens.dev/s/cloudflare+workers      # web search

Preview before you pay (free)

An agent shouldn't have to buy a call to learn what it returns. POST /preview is free and answers that up front — the live price, a one-line summary, and a real sample of the exact response shape:

curl -X POST https://api.weblens.dev/preview \
  -H "Content-Type: application/json" \
  -d '{"endpoint": "/answer"}'

{
  "endpoint": "/answer",
  "price": "$0.05",
  "currency": "USD",
  "summary": "A direct answer with inline [n] citations to real sources.",
  "sample": { "query": "...", "answer": "...", "citations": [], "confidence": 0.92 },
  "sampleType": "recorded",
  "livePreviewAvailable": false,
  "livePreviewHint": "This endpoint calls a paid upstream provider, so free live previews are not offered — the recorded sample shows the exact response shape."
}

Live vs recorded. A real, truncated live preview runs only for endpoints whose marginal cost is a plain fetch — currently /fetch/basic, /contents and /map — and only when you pass a url. Everything else is backed by a metered upstream (SerpAPI, Anthropic); running those free would burn upstream credits, so they return the recorded sample, which still shows every field name and type. An endpoint that isn't sold returns 404.

# live: the first 500 chars of the real result, for free
curl -X POST https://api.weblens.dev/preview \
  -H "Content-Type: application/json" \
  -d '{"endpoint": "/fetch/basic", "url": "https://example.com"}'

Rate limited to 10 requests/hour per IP like the rest of the free tier.

ERC-8004: receipts and feedback

ERC-8004 (Trustless Agents) keeps compact signals on-chain and the detailed documents off-chain. WebLens hosts the off-chain half — the part a service operator can run without deploying a contract. To be precise about what that is and isn't: WebLens is not registered on-chain, holds no agent id, and writes nothing to any registry.

Two more things worth stating plainly:

• A receipt's mac is a symmetric HMAC tag (the same construction as proof-of-context). Only a holder of the key can verify it — it is not a third-party-verifiable signature, and nothing here is trustless.
• The buyer authors the feedback document and posts giveFeedback() themselves. WebLens only hosts the document verbatim and returns its hash; it never authors, edits, or submits feedback.

Endpoints

All paid endpoints are POST with a JSON body. Prices are per request in USDC.

Core

Search

Social

Extraction & research

/research/deep is the only long-running endpoint: it plans sub-questions, runs a web search per sub-question, fetches and dedupes the sources, then synthesizes a cited answer — all in one synchronous call. A standard run (3 sub-questions, 8 sources, $0.20) takes roughly 30–60 seconds; deep (5 sub-questions, 12 sources, $0.35) takes longer. Set a generous HTTP timeout (120s+) and don't retry on a client timeout, or you'll pay twice.

Intelligence

Utility

Free and unauthenticated: /, /health, /docs, /openapi.json, /llms.txt, /discovery, /.well-known/x402, /mcp, /r/{url}, /s/{query}, /free/fetch, /free/search, /preview, /.well-known/agent-registration.json, /receipts/{requestId}, /feedback, /feedback/{id}.

Dynamically-priced endpoints (/fetch/pro, /extract) apply a complexity multiplier — up to 3× for bot-protected domains. Cached responses on the fetch family are 70% cheaper. The exact price is always in the 402 challenge.

Use with AI agents (MCP)

Remote HTTP — nothing to install:

{ "mcpServers": { "weblens": { "url": "https://api.weblens.dev/mcp" } } }

Local with automatic payment:

{
  "mcpServers": {
    "weblens": {
      "command": "npx",
      "args": ["-y", "@weblens/mcp"],
      "env": { "PRIVATE_KEY": "0xYourPrivateKeyHere" }
    }
  }
}

See mcp-server/README.md for the full tool list and setup.

API usage

# Fetch a page as markdown
curl -X POST https://api.weblens.dev/fetch/basic \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

# Search, with page content included
curl -X POST https://api.weblens.dev/search \
  -H "Content-Type: application/json" \
  -d '{"query": "latest AI news", "limit": 5, "includeContent": true}'

# Structured extraction
curl -X POST https://api.weblens.dev/extract \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/product",
       "schema": {"title": {"type": "string"}, "price": {"type": "number"}},
       "instructions": "Extract the product title and price"}'

# Crawl a site (pay for the page budget you request)
curl -X POST https://api.weblens.dev/crawl \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com", "limit": 10, "maxDepth": 2}'

# Deep research with inline citations (slow — 30-60s, so raise the client timeout)
curl --max-time 120 -X POST https://api.weblens.dev/research/deep \
  -H "Content-Type: application/json" \
  -d '{"query": "How are AI agents using micropayments in 2026?", "depth": "standard"}'