Remote sandbox · never local Chrome
Screenshots for
AI agents
Capture real pages in a remote Playwright sandbox. MCP, CLI, and HTTP — without handing agents your cookies or laptop browser.
100 free captures · no card · hard stop when free is empty
Live demo capture via Snapliz API — not a mock.
Agent install surface
MCP · CLI · TypeScript · Python · cURL
Reliability
Agents need pixels, not broken HTML.
Raw fetch fails on JS apps, soft 404s, and 500 pages that look fine until the model invents content. Snapliz renders the real page and returns a stable image plus metadata — even when the site is wrong.
Security
Keep agents out of local Chrome.
Local headless shares bank sessions, Gmail, and SSO. Snapliz runs a cold remote sandbox every time. Cookie dismiss is a toy set of selectors — honest, not a fake 50k CMP claim.
Local headless vs Snapliz
| Concern | Local headless | Snapliz |
|---|---|---|
| JS-rendered pages | DIY waits | Real Chromium |
| Error pages | Often fail | Image + http_status |
| User cookies / SSO | Shared profile risk | Cold sandbox |
| Agent install | Custom scripts | MCP · CLI · SDK |
| Free exhausted | Silent overage? | Hard 402 |
Install
MCP. CLI. SDKs. One contract.
Agent-native like Postiz: one AGENTS.md, first-party MCP, shell CLI. Not a bolt-on marketing page for “OpenClaw workflows.”
MCP
Cursor, Claude Desktop, Claude Code. Tools: capture, full page, usage, formats.
CLI
Shell agents, CI, eval harnesses. Same contract as the HTTP API.
HTTP + SDKs
ScreenshotOne-compatible /api/take. TypeScript and Python clients.
{
"mcpServers": {
"snapliz": {
"command": "npx",
"args": ["-y", "@agentscreenshots/mcp-server"],
"env": {
"AGENTSCREENSHOTS_API_KEY": "asx_live_..."
}
}
}
}Ship agents that see the real web.
Free tier to wire it up. Hard 402 when included captures are gone — until billing is active.