The X API costs $100–$5,000+/month. Hosted scraper services bill you per tweet and hold your data on their servers. Meanwhile, your own browser already shows you everything you need — for free.

x-scraper-no-api turns that browser into a clean data pipeline. You log in once, manually, in a real browser window. After that, one command exports search results, timelines, and threads as LLM-ready JSON, CSV, or Markdown — straight to your disk, never through a third party.

• 🔑 Zero API keys — no developer account, no app approval, no billing page
• 🖐️ Manual login, once — your password and 2FA never touch this tool; the browser session persists like any normal browser
• 🧠 LLM-ready output — JSON, JSONL, CSV, or token-friendly Markdown built for AI agent context windows
• 🤖 Agent-native — ships a ready-to-install Skill for OpenClaw, Hermes, and any CLI-capable agent
• 🎯 Full X search syntax — from:, since:, until:, min_faves:, filter:media, lang: and every advanced operator
• 🧱 Resilient parsing — reads the structured GraphQL data X sends its own frontend instead of scraping fragile HTML class names
• 🐢 Polite by design — human-like pacing, hard item caps, automatic backoff when X signals pressure
• 🔒 Private — everything runs locally; nothing is proxied, relayed, or uploaded anywhere

Three steps. Two minutes.

1. Install Xcode Command Line Tools (needed for native builds):

xcode-select --install

2. Install Node.js — grab the official installer or nvm from nodejs.org/en/download (Node 18 or newer).

3. Install the scraper:

mkdir -p 'xscraper' && cd 'xscraper' && npm install github:JoinArtisanVent/x-scraper-no-api

Then finish setup and log in manually (one time):

npx playwright install chromium
npx xscraper login        # a browser opens — sign into X yourself

That's it. Your session lives in ~/.xscraper/ and every future run is headless.

# Search — full X advanced-search syntax
xscraper search "from:openai since:2026-01-01 min_faves:500" --limit 50 --format md

# The Latest tab instead of Top
xscraper search "ai agents" --latest --limit 25

# A public timeline
xscraper timeline @nasa --limit 100 --format csv -o nasa.csv

# One post + its public replies
xscraper tweet https://x.com/nasa/status/1846987139428634858 --format json

{
  "id": "1846987139428634858",
  "url": "https://x.com/nasa/status/1846987139428634858",
  "created_at": "Wed Oct 16 12:34:56 +0000 2026",
  "text": "Liftoff! …",
  "lang": "en",
  "author": { "username": "nasa", "name": "NASA", "verified": true, "followers": 80000000 },
  "metrics": { "replies": 1200, "reposts": 4800, "likes": 32000, "views": 1500000 },
  "hashtags": ["EuropaClipper"],
  "media": [{ "type": "photo", "url": "https://pbs.twimg.com/media/…" }],
  "scraped_at": "2026-09-26T10:00:00.000Z"
}

This repo ships an agent Skill at skills/x-scraper-no-api/SKILL.md — a portable manifest that teaches agents how (and how not) to use the tool.

• OpenClaw / Hermes: point your agent at the skills/ directory, or copy SKILL.md into your agent's skills folder.
• Any LLM script: shell out to the CLI and feed stdout into your prompt — see examples/agent-workflow.md.

import subprocess
ctx = subprocess.run(
    ["xscraper", "search", "local-first software", "--limit", "50", "--format", "md"],
    capture_output=True, text=True, timeout=600,
).stdout  # ready for your prompt

The Skill enforces a safety contract: public data only, small limits, no credential handling, scraped content treated as untrusted input.

Want to hack on it? Welcome — this project lives or dies by community maintenance.

Architecture (deliberately small — 5 files):

src/
├── cli.js        # Commander CLI
├── session.js    # Persistent Playwright profile + manual login
├── scraper.js    # GraphQL response interception + normalization
├── ratelimit.js  # Human pacing, hard caps, backoff
└── output.js     # JSON / JSONL / CSV / Markdown writers
skills/x-scraper-no-api/SKILL.md   # Agent Skill manifest

How it works: we never call X's internal endpoints directly. A real Chromium instance loads x.com exactly as it does for a human; we passively capture the GraphQL JSON X streams to its own frontend and normalize it into a stable schema. When X redesigns its markup, we don't break — we never read the markup.

Contributing: see CONTRIBUTING.md. The most valuable PRs are parser-robustness fixes, new job types (lists, trends), and fixture tests for the normalizer. PRs adding CAPTCHA bypass, proxy rotation, or credential handling will be declined — compliance is a feature here.

Roadmap

• List and community job types
• Fixture-based test suite for the normalizer
• Scheduled/resumable runs with checkpoints
• MCP server wrapper
• Linux/Windows install one-liners

This tool is built to stay on the right side of the line:

• ✅ Public posts only — protected accounts, DMs, and restricted content are not supported and never will be
• ✅ You authenticate yourself — manual login in your own browser; the tool never sees credentials, cookies, or 2FA codes
• ✅ No circumvention — no CAPTCHA solving, no proxy rotation, no account pooling, no rate-limit evasion
• ✅ Paced like a human — conservative delays, small caps, automatic cooldowns
• 📋 Results may contain personal data — have a lawful purpose, minimize storage, honor deletion requests (GDPR & friends apply)
• ⚖️ You are responsible for complying with X's Terms of Service and applicable law in your jurisdiction

Disclaimer: This is an independent, community-maintained open-source project. It is not affiliated with, endorsed by, or sponsored by X Corp. "Twitter" and "X" are trademarks of X Corp. No X Corp code, assets, or proprietary material is included in this repository.

Do I need an X account? Yes — X removed anonymous browsing years ago. You log in manually once, in your own browser. Use an account you're comfortable browsing with.

Will my account get banned? The tool paces itself conservatively and caps every run. No tool can guarantee zero risk — keep limits modest and don't run it around the clock.

Why 200 items max? Deliberately. This is a research and agent-context tool, not a bulk-harvesting machine. Two focused 50-item searches beat one giant trawl.

It returned fewer results than my limit! The timeline was exhausted — that's normal, not a bug.

Windows/Linux? Everything except the Quickstart wording is cross-platform — install Node 18+, then the same npm install github:… line works everywhere.

Why does login open a real browser? Because that's the point. Manual login means no credential handling, no automation-detection games, and a session X itself issued.