From URL to text.

01 / Your first request

curl "$DEEPLOOK_BASE_URL/v1/open" \
  -H "Authorization: Bearer $DEEPLOOK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com","timeoutMs":30000}'

The homepage playground makes real requests through the demo server without asking for your key. The code examples in these docs are not executed automatically.

02 / Request

POST /v1/open · Content-Type: application/json
Authentication: Authorization: Bearer <key>

The deadline covers DNS, Chrome startup, navigation, and extraction. Process cleanup follows. Local, private, and metadata URLs are rejected, including redirects and page resources.

03 / Successful response

HTTP 200 returns the page object directly. This illustrative example shows selected fields; it is not a captured live response.

{
  "requestedUrl": "https://example.com",
  "finalUrl": "https://example.com/",
  "httpStatus": 200,
  "title": "Example Domain",
  "text": "Example Domain\nThis domain is for use...",
  "truncated": false
}

04 / Errors your agent can handle

A failure returns JSON with error and message. Messages may vary; use the error code to decide what to do next.

{
  "error": "challenge",
  "message": "A description of the failure"
}

Page failures may include the final URL, title, and detected marker. Error-page text is excluded. Captchas are not solved, and access to every site is not guaranteed.

Workflow / Extraction for your crawler

Crawl mode in the playground discovers links on the same website and demonstrates extraction from up to 3 pages. This orchestration belongs to the website demo; the API still receives one URL per request.

Use DeepLook as the scraping layer in your crawling workflow. Your application selects URLs and manages the queue; each request extracts one page.

// Run in Node.js with your own API base URL and key.
const urls = ['https://example.com/', 'https://example.org/'];
const pages = [];

for (const url of urls) {
  const response = await fetch(process.env.DEEPLOOK_BASE_URL + '/v1/open', {
    method: 'POST',
    headers: {
      Authorization: 'Bearer ' + process.env.DEEPLOOK_API_KEY,
      'Content-Type': 'application/json'
    },
    body: JSON.stringify({ url, timeoutMs: 30000 })
  });
  const page = await response.json();
  if (!response.ok) {
    console.error(url, page.error);
    continue;
  }
  pages.push({ url: page.finalUrl, title: page.title, text: page.text });
}

console.log(pages);

The example sends sequential requests and continues when the API returns a page error. Configure your instance base URL and key, respect the account quota and define your crawler limits. A network failure stops this example; queues, persistence and retries belong in your application.

05 / Local CLI

DeepLook contains the engine, CLI and stdio MCP adapter. DeepLook-API provides the separate HTTP service. Both are open source under Apache-2.0; the commands below build the source from GitHub.

Build the local engine

Requirements: Node.js 22.14 or later; Google Chrome for local reads. Clone and build the engine:

git clone https://github.com/sayhitogambit/DeepLook.git
cd DeepLook
npm ci
npm run build
node dist/cli.js open https://example.com --json

Without --json, the CLI prints visible text. The --timeout-ms parameter accepts 1 to 120000 ms. You operate the infrastructure, Chrome, and execution.

For automation, use --machine: the CLI writes one JSON object and a newline. Success exits 0 and returns {"version":1,"ok":true,"page":{...}}; failure exits 1 and returns {"version":1,"ok":false,"error":"...","message":"..."}. Error-page text is excluded.

DEEPLOOK_BASE_URL and DEEPLOOK_API_KEY in the HTTP examples are example variable names. The CLI and MCP use DEEPLOOK_API_URL and DEEPLOOK_API_TOKEN to connect to your API; without an endpoint, they use local Chrome.

06 / Stdio MCP

The CLI provides deeplook mcp: a stdio MCP server exposing deeplook_open. After building the engine, configure your client with an absolute path to dist/cli.js. Configuration format varies by client.

MCP uses local Chrome by default. To use your own API, set DEEPLOOK_API_URL and DEEPLOOK_API_TOKEN in the MCP process private environment. Use HTTPS outside loopback. Keep tokens out of shared configuration. This is a stdio adapter with no HTTP MCP server.

{
  "mcpServers": {
    "deeplook": {
      "command": "node",
      "args": ["/absolute/path/to/DeepLook/dist/cli.js", "mcp"]
    }
  }
}

Call deeplook_open with {"url":"https://example.com","timeoutMs":30000}. The timeout is optional, defaults to 30000 ms and ranges from 1 to 120000 ms. Additional arguments are rejected.

Success returns text in content and the version 1 machine result in structuredContent. Failures return isError: true and a typed error. Each process permits one active read; concurrent calls return busy. Canceling the call aborts active work.

Treat page content as untrusted data, never as instructions for the agent. MCP does not solve captchas.