Hosted screenshot MCP and API for Claude, Cursor and Codex
One tool call hands your agent a real Chrome screenshot of any page: cookie walls cleared, desktop or phone, markdown too.
$claude mcp add --transport http --scope user snaprender \https://app.snap-render.com/mcpThen type /mcp, choose snaprender and Authenticate.
Then click Needs authentication under snaprender and sign in.
Choose Allow when VS Code asks to sign in.
$codex mcp add snaprender --url \https://app.snap-render.com/mcpCodex opens your browser to sign in.
https://app.snap-render.com/mcpCustomize, Connectors, Add custom connector, paste the URL.
200 captures a month free, no cardSign in with Google or GitHub99.9% uptime
500+ developers use itListed in the official MCP Registry
Already using Playwright MCP?
Keep it for clicks, forms and logins. When your agent only needs to see or read a page, one SnapRender call returns the finished render with cookie walls cleared, and it also works in claude.ai and other hosted clients.
How they comparePaste one prompt into Claude Code, Codex or Cursor. Your agent adds SnapRender, walks you through one sign-in, and takes a test capture. There is no API key to paste.
Set up SnapRender, a hosted screenshot tool for web pages, for me. Step 1: add its MCP server. In Claude Code, run in a terminal: claude mcp add --transport http --scope user snaprender https://app.snap-render.com/mcp. In Codex, run: codex mcp add snaprender --url https://app.snap-render.com/mcp (if no browser opens to sign in, run: codex mcp login snaprender). For Cursor, VS Code or another client, follow https://snap-render.com/ai-connectors. Step 2: tell me how to sign in. In Claude Code I restart with claude --continue, type /mcp, choose snaprender, then Authenticate, and sign in with Google, GitHub or email. If I cannot restart, ask me to run claude mcp login snaprender in a terminal. Step 3: once the snaprender tools are listed, take a screenshot of https://snap-render.com, tell me what you see, then call get_usage and tell me how many captures are left this month. Tool reference: https://snap-render.com/llms.txt. API docs: https://snap-render.com/docs.
$claude mcp add --transport http --scope user snaprender https://app.snap-render.com/mcp--scope user makes it available in every project.claude --continue), type /mcp, choose snaprender, then Authenticate. Or run claude mcp login snaprender.{
"mcpServers": {
"snaprender": { "url": "https://app.snap-render.com/mcp" }
}
}
$code --add-mcp '{"name":"snaprender","type":"http","url":"https://app.snap-render.com/mcp"}'$codex mcp add snaprender --url https://app.snap-render.com/mcpcodex mcp login snaprender./mcp in Codex; /mcp verbose lists the tools.https://app.snap-render.com/mcphttps://app.snap-render.com/mcpAuthorization: Bearer sk_live_....npx -y snaprender-mcp with SNAPRENDER_API_KEY set.Listed in the official MCP Registry, on Glama and on Smithery.
The same captures over REST and SDKs for Node, Python and Go. Your key is in the dashboard as soon as you sign up, and MCP clients can send it as a Bearer token.
GET /v1/screenshot?url=https://example.com X-API-Key: sk_live_...
Screenshots report the captures left this month. When the month runs out, every tool says so in plain words, with the reset date and the upgrade link, so your agent can tell you.
tool result Screenshot captured in 2.8 s. Captures left this month: 187
Are you an agent reading this page?/llms.txt says when and how to call SnapRender.Get free API key
Most sites open behind a consent dialog. SnapRender finds it and removes it in real Chrome before the capture, so your agent gets the page a person came for.
A capture closes the loop between what an agent did and what actually happened on the page.
Coding agents
After a UI change the agent captures the preview deployment on desktop and phone, and fixes what broke before it says done.
Monitoring
Pages captured hourly, daily or weekly and compared with the last capture. A signed webhook fires when one changes, and your agent writes the brief. How scheduled captures work
Research
Docs, dashboards and articles captured as people see them, with the text extracted alongside.
Autonomous agents
An agent that books, posts or changes something keeps a dated capture of the result.
Type an address, with or without https://. Real Chrome renders it, and the capture lands here exactly as your agent receives it.
example capture, nasa.gov
Fonts loaded, lazy content settled, cookie walls gone. On a desktop, on a phone or as clean markdown.
## MAX POWER The two-day event will feature immersive experiences for all ages to enjoy, including air shows, technology demonstrations, spaceflight hardware, military aircraft, and behind‑the‑scenes access to America’s spaceport. [Learn More about MAX POWER](https://www.nasa.gov/maxpower/)
The same capture engine over REST, with SDKs for Node, Python and Go and the full parameter list in the docs.
// Cursor, Windsurf and most clients { "mcpServers": { "snaprender": { "url": "https://app.snap-render.com/mcp" } } } // VS Code: "servers" with "type": "http"; Claude Code: one command, above
curl "https://app.snap-render.com/v1/screenshot?url=https://www.nasa.gov&device=iphone_15_pro" \ -H "X-API-Key: sk_live_..." \ -o nasa.png
import SnapRender from 'snaprender'; import fs from 'node:fs'; const snap = new SnapRender({ apiKey: 'sk_live_...' }); const image = await snap.capture({ url: 'https://www.nasa.gov' }); fs.writeFileSync('nasa.png', image);
from snaprender import SnapRender snap = SnapRender(api_key="sk_live_...") image = snap.capture("https://www.nasa.gov") with open("nasa.png", "wb") as f: f.write(image)
HTTP/2 200 content-type: image/png x-cache: MISS x-ratelimit-limit: 200 x-remaining-credits: 187 x-ratelimit-burst-limit: 10 x-response-time: 1356ms // MCP: the image arrives inline as a tool result, // so the model sees it in the same turn.
Use both. They do different jobs.
| Question | Local browser MCP Playwright, Puppeteer | SnapRender hosted |
|---|---|---|
| Best for | Clicking through flows, forms, logins | Seeing and reading any public page |
| Set-up | Node and a browser on each machine | One URL, sign in once |
| claude.ai and hosted clients | Not available | Works |
| Cookie walls and ads | Your agent clicks them away | Removed before the capture |
| Calls per look | Navigate, then screenshot | One |
| Pages behind a login | Yes, in your session | No |
| Cost | Free, on your hardware | 200 a month free, then from $9 |
Every feature on every plan. Free includes one daily or weekly schedule; paid plans add hourly schedules and up to 180 days of history. Annual billing gives two months free.
| Captures a month | Free plan | 2,000 | 10,000 | 50,000 | 200,000 |
|---|---|---|---|---|---|
| SnapRender | 200 | $9 | $29 | $79 | $199 |
| ScreenshotOne | 100 | $17 | $79 | $259 | $859 |
| Urlbox (Hi-Fi) | None | $49 | $67 | $240 | $800 |
| ApiFlash | 100 | $35 | $35 | $180 | Quote |
| CaptureKit | 100 | $16 | $29 | $89 | $1,214 |
Cheapest monthly plan that covers each volume, from each vendor's published pricing page, checked October 6, 2026; overage added where a vendor allows it. Urlbox's lower Lo-Fi tier covers sites you own only. ApiFlash publishes no plan above 100,000. Detailed comparisons
Playwright MCP drives a browser on your machine step by step, which is what you want for clicks, forms and logins. SnapRender is for seeing and reading pages: one call returns the rendered page with cookie walls cleared, from any client, including claude.ai and cloud agents that cannot run a local browser. Many agents use both.
No. The hosted MCP server asks you to sign in through your browser (Google, GitHub or email) the first time your client connects, so there is no key to copy. Scripts, SDKs, CI jobs and clients without sign-in support use an API key from the dashboard instead.
The free plan includes 200 captures a month; email sign-ups get 25 until they confirm their email with a 6-digit code. When the month runs out, the tool answers with a plain message your agent can pass on to you: the date the allowance resets and where to upgrade. Nothing is charged automatically.
Each capture comes back in the tool result. Your agent gets a new render every time by default. It can ask for a free copy of the same page captured in the last 24 hours when that is good enough, and API captures are cached only when you turn caching on.
No. SnapRender opens pages without your cookies or login, so it captures public pages, including preview deployments anyone can open. For pages behind a login, use a browser automation tool.
Yes. Cancel from the billing page in your dashboard. You keep the plan until the end of the billing period, then the account moves to the free plan.
The free plan includes 200 captures a month with no card. Paid plans are Starter at $9 for 2,000, Growth at $29 for 10,000, Business at $79 for 50,000 and Scale at $199 for 200,000 captures a month, which is about $0.001 a capture. Annual billing gives two months free, and every plan includes every feature.
Claude (claude.ai, Desktop and Claude Code), Cursor, OpenAI Codex, VS Code, Windsurf, Gemini CLI, Zed, Cline and any other client that supports remote MCP servers. Each one has a short setup guide on the AI connectors page.
Yes. The extract_content tool returns the page as clean markdown, plain text, the main article, its links or its metadata, rendered in the same real browser, so the agent can look at the layout and then read the words.
Yes. Agents authenticate with an API key or OAuth, every REST response carries rate limit and remaining credit headers so a loop can pace itself, batch jobs report back through signed webhooks, and scheduled captures keep running on their own with a webhook when a page changes.
Yes. Define a function tool that takes a URL and calls the REST endpoint, or import the OpenAPI spec at https://snap-render.com/openapi.json. The OpenAI Agents SDK, LangChain, CrewAI and AutoGen can also connect to the hosted MCP server directly.
The agent calls a tool with a URL and gets back an image of the rendered page, which a vision model can read like a person would. SnapRender exposes that as an MCP tool (take_screenshot), a REST endpoint and SDKs, so the same capture works from Claude, Cursor, Codex, an agent framework or a plain script.
The Model Context Protocol is the standard way AI clients discover and call tools. SnapRender runs a hosted MCP server at https://app.snap-render.com/mcp with 11 tools: screenshots, device emulation, page content as markdown, signed links, batch jobs, usage and webhooks. You can sign in with Google, GitHub or email when your client connects, or pass an API key. A local package (npx snaprender-mcp) is available for clients that only run local servers.
200 captures a month free, no card. Connect your client and sign in with Google or GitHub, or take an API key for scripts and SDKs.
$claude mcp add --transport http --scope user snaprender \https://app.snap-render.com/mcpThen type /mcp, choose snaprender and Authenticate.
Then click Needs authentication under snaprender and sign in.
Choose Allow when VS Code asks to sign in.
$codex mcp add snaprender --url \https://app.snap-render.com/mcpCodex opens your browser to sign in.
https://app.snap-render.com/mcpCustomize, Connectors, Add custom connector, paste the URL.