Grok Scraper

Execute queries to Grok AI via Playwright browser automation without requiring an X API KEY. Use when the user wants to "ask Grok", search X for real-time in...

Preview

<video src="./assets/grok-2026-03-15T10-01-45.webm" controls width="100%"></video>

Agent Context: This is a zero-cost alternative to official X APIs. It uses a real browser session (Playwright) via an X Premium account. ALWAYS use this skill when the user wants to query Grok but does not have or want to use an X API KEY.

Prerequisites

• OpenClaw must be installed on the host machine.
• A display/GUI environment is required. This skill launches a real browser window for login. It cannot run on headless cloud servers (no screen). It must be used on a local machine or a remote desktop with a display.
• The user must be logged in to x.com via the browser session saved by npm run login. Without a valid session, all queries will fail.

First-Time Setup

Run these commands once after cloning the repo, before doing anything else:

cd scripts
npm install
npx playwright install chromium

Then log in to x.com to create a session:

npm run login
# A browser window will open — log in to x.com manually, then return to the terminal and press Enter

The session/ directory will be created automatically after a successful login.

Workflow

Step 1: Check Login State

• If session/ directory does not exist: stop and ask the user to run cd scripts && npm run login.
• If it exists: proceed.

Step 2: Execute Query

scripts/run.sh "The user's detailed prompt"

run.sh handles logging, automatic retry on Grok service errors, and login-expiry detection. It is the canonical entry point for all queries.

Step 3: Read Output

• Exit Code 0 → read output/latest.md and present the result.
• Other exit codes → see Error Handling below.

Error Handling

Exit Code	Meaning	Action
0	Success	Read output/latest.md
2	Session expired	Ask user to run cd scripts && npm run login
3	Grok service error	run.sh already retried once; report failure to user
1	Extraction failed	Check if output/debug-dom.json was written → if yes, DOM selectors may have broken — see dom-selector-fix.md

DOM Selectors Breaking

Twitter/X redeploys its front-end regularly, which changes the CSS class names this scraper relies on. If extraction fails with Method: none, follow the fix guide:

→ dom-selector-fix.md

Examples

Standard query

scripts/run.sh "Search for the latest AI news and format as markdown"
# → read output/latest.md

Session expired

1. Run scripts/run.sh → Exit Code 2
2. Tell user: "Session expired, please run cd scripts && npm run login"

DOM selectors broken

1. Run scripts/run.sh → Exit Code 1, output/debug-dom.json exists
2. Follow dom-selector-fix.md to identify new classes and update SELECTORS in scripts/scrape.js

---

Debugging

When diagnosing scraper issues directly, use the bare command — it skips logging and retry logic, making failures easier to inspect.

Flag	Example	Description
(none)	npm run scrape	Run with default prompt
"prompt"	npm run scrape -- "Your question"	Custom prompt
--record	npm run scrape -- --record	Record video to output/grok-<timestamp>.webm
--record <path>	npm run scrape -- --record out.webm	Record video to custom path (relative → output/)
--size WxH	npm run scrape -- --record --size 1920x1080	Set recording resolution (default: 1280x800)

All flags can be combined:

cd scripts
npm run scrape -- "Your prompt" --record --size 1920x1080

When --record is active, the browser runs in headed mode (visible window) with slowMo: 50ms; without it, headless mode is used.