Grabbit
Grabbit gives AI agents eyes on the web. Send a public URL, get a pixel-perfect hosted image.
Installation
Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).
claude mcp add --transport http grabbit 'https://mcp.grabbit.live'
{
"mcpServers": {
"grabbit": {
"type": "http",
"url": "https://mcp.grabbit.live"
}
}
}
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Screenshots as a service, built for AI agents
Point your agent at a URL and get back a hosted image, with no local headless browser to run. Grabbit handles cookie and consent banners, JavaScript-heavy and full-page captures, and many sites that block a local browser.
| Endpoint | https://mcp.grabbit.live (Streamable HTTP) |
| Auth | OAuth 2.1 (claude.ai connectors work out of the box) or a Bearer API key |
| Registry name | live.grabbit/screenshots |
| Setup guides | https://grabbit.live/agents |
Tools
| Tool | What it does |
|---|---|
grab | Screenshot any public URL and get a hosted image back (inline image + remaining credits) |
get_grab | Fetch a previous capture by id |
list_grabs | Browse your recent captures |
get_usage | Credits, plan, and 30-day usage |
get_pricing | Pricing plus an honest, dated comparison against ScreenshotOne, Urlbox, Browserless, and others |
Why a hosted screenshot tool
Most agents can already drive a headless browser. It works on simple pages and quietly fails on the ones that matter: cookie and consent walls, bot detection, login and paywall gates, and heavy client-rendered pages. Grabbit runs the capture on hosted infrastructure, so the shot comes back clean and the agent needs no browser of its own.
Honest bound: no service captures literally every site. A few hard targets behind aggressive IP blocking still resist. Private and internal addresses (including localhost) are blocked by SSRF protection.
Quick start
Claude Code
claude mcp add --transport http grabbit https://mcp.grabbit.live
Then run /mcp and authenticate in the browser (OAuth, no key to paste).
Guide: https://grabbit.live/agents/claude-code
Cursor
~/.cursor/mcp.json (commit a project-scoped .cursor/mcp.json to share with your team):
{ "mcpServers": { "grabbit": { "url": "https://mcp.grabbit.live" } } }
Guide: https://grabbit.live/agents/cursor
OpenAI Codex
~/.codex/config.toml:
[mcp_servers.grabbit]
url = "https://mcp.grabbit.live"
bearer_token_env_var = "GRABBIT_API_KEY"
Guide: https://grabbit.live/agents/codex
OpenClaw
Add the hosted MCP server, or wrap the one-line CLI in a skill. Guide: https://grabbit.live/agents/openclaw
Cursor Marketplace
Install Grabbit directly from the Cursor Marketplace (when listed), or add manually:
~/.cursor/mcp.json (or project-scoped .cursor/mcp.json):
{ "mcpServers": { "grabbit": { "url": "https://mcp.grabbit.live" } } }
This repo includes Cursor plugin packaging (.cursor-plugin/plugin.json, mcp.json, skill) for marketplace submission.
Agent Skills
This repo ships a portable skill folder (skills/grabbit-screenshots/) that teaches agents when and how to use the Grabbit hosted screenshot MCP.
Install via skills.sh
npx skills add BrainGridAI/grabbit-mcp --skill grabbit-screenshots
After public installs, the skill can appear on skills.sh via install telemetry (no separate submit form).
Publish to ClawHub
From the repo root:
clawhub skill publish ./skills/grabbit-screenshots --slug grabbit-screenshots --name "Grabbit Screenshots" --version 1.0.0
Install via OpenClaw / Hermes
The skill is published on ClawHub at https://clawhub.ai/braingrid/skills/grabbit-screenshots
openclaw skills install @braingrid/grabbit-screenshots
# or
clawhub install grabbit-screenshots
Hermes can install the same ClawHub skill (ClawHub is a Hermes skill source) and connect MCP at https://mcp.grabbit.live.
CLI
npx grabbit.live login # browser sign-in, keys saved locally
npx grabbit.live nyt.com # screenshot a URL, hosted image back
npx grabbit.live mcp # print MCP config for your agent
Pricing
Flat $0.002 per live capture. Prepaid credits that never reset or expire. Test keys return free placeholders so you can wire it up before paying. Full pricing and the comparison table: https://grabbit.live/#pricing
Links
- Website: https://grabbit.live
- Agent setup guides: https://grabbit.live/agents
- API reference: https://grabbit.live/screenshot-api
- LLM-readable docs: https://grabbit.live/llms.txt
The server.json in this repo is the manifest published to the official MCP
Registry. Grabbit's source lives in a private repo; this is the public home for
the MCP server's manifest and docs.
Questions fréquentes
Qu'est-ce que Grabbit ?
Grabbit gives AI agents eyes on the web. Send a public URL, get a pixel-perfect hosted image.
Comment installer Grabbit ?
Il fonctionne comme serveur distant. Dans Claude Code : claude mcp add --transport http grabbit 'https://mcp.grabbit.live'. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.
Faut-il une clé API pour Grabbit ?
Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.
Grabbit est-il open source ?
Oui. Son code est sur GitHub sous licence MIT.
Qui maintient Grabbit ?
Il est publié par BrainGridAI. AI.tn le référence sans le maintenir ni le recommander.
Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.