Skip to content
AgentSearch

Native x402 on Base · also on Pocket Network

Headless browser rendering API for AI agents

AgentSearch Web Render loads a page in headless Chromium, lets its JavaScript run, and returns the title, rendered text/markdown and HTML. It is a JavaScript rendering API for AI agents: render SPA to text, read dashboards and client-rendered lists, and do dynamic page scraping politely.

$0.005 per call (direct x402)Free trial: 3 calls/dayUSDC on Base via x402No account, no API keyPocket MainNet · Beta TestNetPOST /v1/renderGET /v1/healthGET /v1/version

Try Web Render free (no key, no wallet)

Every IP gets 3 free render calls per day on the AgentSearch relay, with no account, API key or wallet (text output: markdown, text, HTML and links; screenshots are paid only). Send a public http(s) URL; private, internal and cloud-metadata addresses are refused. Each answer includes a free_trial object with remaining_today. Past the trial the relay answers HTTP 429 free_trial_limit_reached with a link to wallet setup; paid calls are $0.005 each on the direct x402 API, with no daily cap.

Try it in your browser

3 free render calls per day per IP. No signup. Page text is shown as plain text.

Free trial over HTTP

curl -s https://relay.agentsearchhq.com/v1/render \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com"}'

Keep going after the trial

Make a new wallet, add $1 of USDC on Base ($1 covers 100 calls), and set AGENTSEARCH_WALLET_PRIVATE_KEY in the MCP server or your framework package. No ETH is needed for gas. Step by step: wallet setup in about 2 minutes. The MCP server (1.4.0+), langchain-agentsearch (0.3.0+) and the n8n node (0.3.0+) use this free trial automatically when no wallet key is set; other framework packages still need the wallet key for render.

Use Web Render from an MCP client

The recommended MCP server is npx -y @agentsearchhq/agentsearch-mcp. It adds agentsearch_render to Claude Desktop, Cursor, VS Code, Windsurf or any stdio MCP host. Web Render is a paid tool: add a wallet key and each call pays $0.005 in USDC on Base straight to AgentSearch over x402, with no account or API key. Search in the same server is free with no key (up to 25 searches per day per IP).

Claude Desktop or Cursor config

{
  "mcpServers": {
    "agentsearch": {
      "command": "npx",
      "args": ["-y", "@agentsearchhq/agentsearch-mcp"],
      "env": {
        "AGENTSEARCH_WALLET_PRIVATE_KEY": "0xYOUR_NEW_WALLET_KEY"
      }
    }
  }
}

Spending limits built in

The server never signs more than $0.005 per call. It stops at a session cap of $1.00 by default (AGENTSEARCH_MAX_TOTAL_ATOMIC, in USDC atomic units). It signs only USDC on Base to AgentSearch's pinned payee (or the Pocket portal's, if you pick the Pocket route). Use a new wallet holding only what you are willing to spend. No ETH is needed for gas. Without a key, agentsearch_render makes no network call and returns setup instructions.

Also on Pocket Network: Pocket's Agentic Portal MCP, npx -y @pocket-network/agentic-portal-mcp, with call_service on agentsearch-web-render-v1 and path /v1/render. See pricing for what is and isn't charged.

Use Web Render from your agent framework

Every AgentSearch package ships agentsearch_render next to free search. Set AGENTSEARCH_WALLET_PRIVATE_KEY to a new wallet holding a little USDC on Base and it pays $0.005 per call ($1 covers 100 calls); without a key it returns setup instructions and makes no call. Use it when Extract comes back empty (JavaScript apps); the Python and JS LangChain packages and CrewAI can retry thin extracts with render automatically (render_fallback).

FrameworkInstallTool
LangChain (Python)pip install "langchain-agentsearch[paid]"AgentSearchRenderTool()
LangChain.jsnpm install @agentsearchhq/langchainAgentSearchRenderTool() or agentSearchTools()
CrewAIpip install "crewai-agentsearch[paid]"AgentSearchRenderTool()
LlamaIndexpip install "llama-index-tools-agentsearch[paid]"AgentSearchToolSpec().to_tool_list()
OpenAI Agents SDKpip install "openai-agents-agentsearch[paid]"agentsearch_render
LangGraph, Google ADK, Pydantic AI, Vercel AI SDKthe MCP server aboveagentsearch_render
n8nAgentSearch node n8n-nodes-agentsearch 0.3.0+, or MCP Client Tool + the MCP server aboveRender Page operation (agentsearch_render via MCP)

Search, then read: read JavaScript pages from an agent · scrape JS-heavy pages · LangChain render fallback · extract vs render.

What it does

Real browser rendering

Headless Chromium with a fresh browser context per request. Choose formats from markdown, text, html and links, and wait with wait_until or wait_for_selector.

Optional screenshot

Set screenshot to get a PNG or JPEG back as base64. Full-page captures are capped at 4,000 px and the image at 1 MB.

Polite by design

robots.txt is honored (token AgentSearchRender). No logins, clicks or form submission. Bot walls are reported as TARGET_BOT_WALL, not bypassed.

When to use it vs. the other two

AgentSearch has three services. Pick by what your agent has in hand and where the content lives.

Find sources

Web Search

You have a question, not a URL. Get up to 5 ranked results with title, URL, snippet, score, domain and date.

Web Search API →
Read a static page

Web Extract

You have a URL and the content is in the HTML the server sends: docs, blogs, news, most marketing pages. Fast, with no browser overhead.

Web Extract API →
This page

Web Render

The content only appears after JavaScript runs: SPAs, dashboards, client-rendered lists. Headless Chromium renders it first.

When to use Render

Pick the tool by what you have. Start with Extract, and call Render once only when the result is thin. Don't retry Extract on the same URL.

No URL yet?

Use Search

Find pages first with Search.

Text is in the HTML?

Use Extract

A URL that serves its text in HTML: use Extract to get clean markdown.

Needs JavaScript?

Use Render

The content only shows after JavaScript runs. Render loads the page in headless Chromium and returns markdown, text, HTML, links and an optional screenshot.

Signs you need Render

  1. Extract returned only a title, or a few hundred characters at most.
  2. The text says “enable JavaScript” or “requires JavaScript”.
  3. The page is a single-page app whose body is an empty <div id="root">.
  4. Content appears only after the page loads data in the browser.

robots.txt and price

robots.txt: Render respects robots.txt. If a site blocks bots, Render refuses before fetching, and you are not charged (ROBOTS_DISALLOWED 422 when called directly, UPSTREAM_REJECTED 400 through the portal). Pick another source.

Price: $0.005 per call on the direct x402 API (USDC on Base), charged only when content comes back. No signup. An Extract followed by a Render costs $0.007. Also on Pocket Network at $0.005.

Python (Pocket call_service wrapper; or use any x402 client on the direct API)

page = call_service(
    "agentsearch-web-render-v1",
    "/v1/render",
    {"url": url, "formats": ["markdown"], "max_chars": 8000},
)  # Pocket route: POST https://agent.pocket.network/v1/agentsearch-web-render-v1/v1/render
# Direct x402 (recommended): POST https://api.agentsearchhq.com/x402/v1/render with {"url": url}

See the Extract-then-Render fallback in Python →

Request and response

Request body (POST /v1/render)

{
  "url": "https://example.com/",
  "formats": ["text"],
  "max_chars": 500
}

# Other options from the spec:
#   "formats": ["markdown", "text", "html", "links"]
#   "screenshot": true, "screenshot_format": "png" | "jpeg"
#   "wait_until": "domcontentloaded" | "load" | "networkidle"
#   "wait_for_selector": "#app"
#   "viewport": {"width": 1280, "height": 800}

Response shape (from the spec)

{
  "request_id": "<string>",
  "url": "https://example.com/",
  "requested_url": "https://example.com/",
  "status_code": 200,
  "title": "Example Domain",
  "meta": {
    "render_ms": <integer>, "fetched_at": "<date-time>",
    "chars": <integer>, "truncated": <boolean>,
    "js": <boolean>, "robots": "<string|null>",
    "blocked_requests": <integer>,
    "content_type": "<string|null>", "lang": "<string|null>"
  },
  "error": null,
  "warnings": [],
  "links": [ ... ],
  "text": "<rendered text>",
  "markdown": "<rendered markdown>",
  "html": "<only if requested>",
  "screenshot": { "format", "width", "height", "bytes", "base64" }
}

Field names come from the render OpenAPI spec; values in angle brackets are placeholders. Target-site failures (for example TARGET_DEADLINE on the 4.2 s render deadline) return HTTP 200 with error {code, message, retryable}; bad input returns 4xx.

How to connect

HTTP (recommended)

Direct x402 API (pay per call)

POST https://api.agentsearchhq.com/x402/v1/render. An unpaid request returns 402 with the terms in PAYMENT-REQUIRED; any x402 client (@x402/fetch, Python x402) pays $0.005 in USDC on Base and retries. Settled only after a successful answer. No account, no API key.

Direct x402 docs →
Also on Pocket Network

Pocket Agentic Portal

POST https://agent.pocket.network/v1/agentsearch-web-render-v1/v1/render at $0.005 per call over x402 (Base) or MPP (Tempo).

Open the portal page →
Pocket relays

Call it on Pocket Network

Web Render is served by AgentSearch suppliers on Pocket MainNet (registered 2026-10-02) and Beta TestNet as agentsearch-web-render-v1. Gateways configure it as a REST passthrough service.

View on Pocket Explorer →

Use cases

Single-page apps

Read React, Vue or Angular apps whose HTML is an empty shell until JavaScript runs.

Dashboards and listings

Capture client-rendered tables, product lists and status pages as text or markdown for an agent.

Visual checks

Grab a screenshot alongside the text when an agent needs to see the page layout.

FAQ

What does the headless browser rendering API return?

The final URL, HTTP status, title, rendered text and markdown, links, and on request the rendered HTML and a PNG/JPEG screenshot, plus meta such as render time and character count.

When should I use Render instead of Extract?

Use Web Extract for static pages: it is faster. Use Render when the content only appears after JavaScript runs, or when you need a screenshot.

How do I call it today?

From an MCP client, add npx -y @agentsearchhq/agentsearch-mcp with AGENTSEARCH_WALLET_PRIVATE_KEY set and call agentsearch_render; it pays $0.005 per call in USDC on Base straight to AgentSearch. Over HTTP, POST https://api.agentsearchhq.com/x402/v1/render with any x402 client (docs). An unpaid request returns 402 with the payment terms; the client pays $0.005 in USDC on Base and retries. Also on Pocket Network: POST https://agent.pocket.network/v1/agentsearch-web-render-v1/v1/render ($0.005). No account, no API key. On Pocket Network, Pocket's Agentic Portal MCP also works, with call_service on agentsearch-web-render-v1. The contract is in agents.md.

Does it log in or bypass bot protection?

No. It honors robots.txt, never logs in, clicks or submits forms, and reports bot walls as TARGET_BOT_WALL instead of bypassing them.

How long can a render take?

The render deadline is 4.2 seconds. A slow page returns what loaded so far with error.code TARGET_DEADLINE.

Which health endpoints are there?

GET /v1/health (readiness), GET /v1/version (identity) and GET /v1/capabilities (limits, formats and error codes).

Other AgentSearch services

Guides: Search → Extract → Render: when to step up to rendering · Error handling for AI agent web tools