Index of every page on this site: llms.txt
All skills
vercel-labs avatar

/vercel-sandbox

@5aad200 official
by Vercel Labsvercel-labs/agent-browser53k stars
3,714
Run it yourself

Run agent-browser + Chrome inside Vercel Sandbox microVMs for browser automation from any Vercel-deployed app. Use when the user needs browser automation in a Vercel app (Next.js, SvelteKit, Nuxt, Remix, Astro, etc.), wants to run headless Chrome without binary size limits, needs persistent browser sessions across commands, or wants ephemeral isolated browser environments. Triggers include "Vercel Sandbox browser", "microVM Chrome", "agent-browser in sandbox", "browser automation on Vercel", or any task requiring Chrome in a Vercel Sandbox.

  • 1 file
  • 8.4 KB
  • Updated 3 weeks ago
  • GitHub
Use this Skill: https://skilld.dev/gh/vercel-labs/agent-browser/vercel-sandbox

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈141 tokens for metadata: the name and description. ≈2k when used: this file.

Description uses 7.0% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈563 characters, or 7.0%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Browser Automation with Vercel Sandbox

Run agent-browser + headless Chrome inside ephemeral Vercel Sandbox microVMs. A Linux VM spins up on demand, executes browser commands, and shuts down. Works with any Vercel-deployed framework (Next.js, SvelteKit, Nuxt, Remix, Astro, etc.).

Dependencies

pnpm add @agent-browser/sandbox @vercel/sandbox

The sandbox VM needs system dependencies for Chromium plus agent-browser itself. The @agent-browser/sandbox helpers install them by default for fresh sandboxes and use sandbox snapshots (below) for sub-second startup. Pass installSystemDependencies: false only when the sandbox image already provides Chromium's required libraries.

Core Pattern

import {
  createAgentBrowserSnapshot,
  runAgentBrowserCommand,
  withAgentBrowserSandbox,
  type VercelSandboxSession,
} from "@agent-browser/sandbox/vercel";

async function withBrowser<T>(
  fn: (sandbox: VercelSandboxSession) => Promise<T>,
): Promise<T> {
  return withAgentBrowserSandbox(fn);
}

Screenshot

The screenshot --json command saves to a file and returns the path. Read the file back as base64:

export async function screenshotUrl(url: string) {
  return withBrowser(async (sandbox) => {
    await runAgentBrowserCommand(sandbox, ["open", url]);

    const titleResult = await runAgentBrowserCommand<{ data?: { title?: string } }>(sandbox, [
      "get", "title",
    ]);
    const title = titleResult.json?.data?.title || url;

    const ssResult = await runAgentBrowserCommand<{ data?: { path?: string } }>(sandbox, [
      "screenshot",
    ]);
    const ssPath = ssResult.json?.data?.path;
    if (!ssPath) throw new Error("Screenshot did not return a file path.");
    const b64Result = await sandbox.runCommand("base64", ["-w", "0", ssPath]);
    const screenshot = (await b64Result.stdout()).trim();

    await runAgentBrowserCommand(sandbox, ["close"], { json: false });

    return { title, screenshot };
  });
}

Accessibility Snapshot

export async function snapshotUrl(url: string) {
  return withBrowser(async (sandbox) => {
    await runAgentBrowserCommand(sandbox, ["open", url]);

    const titleResult = await runAgentBrowserCommand<{ data?: { title?: string } }>(sandbox, [
      "get", "title",
    ]);
    const title = titleResult.json?.data?.title || url;

    const snapResult = await runAgentBrowserCommand(sandbox, ["snapshot", "-i", "-c"], {
      json: false,
    });

    await runAgentBrowserCommand(sandbox, ["close"], { json: false });

    return { title, snapshot: snapResult.stdout };
  });
}

Multi-Step Workflows

The sandbox persists between commands, so you can run full automation sequences:

export async function fillAndSubmitForm(
  url: string,
  data: Record<string, string>,
  postSubmitWaitArgs: string[],
) {
  return withBrowser(async (sandbox) => {
    await runAgentBrowserCommand(sandbox, ["open", url]);

    const snapResult = await runAgentBrowserCommand(sandbox, ["snapshot", "-i"], {
      json: false,
    });
    const snapshot = snapResult.stdout;
    // Parse snapshot to find element refs...

    for (const [ref, value] of Object.entries(data)) {
      await runAgentBrowserCommand(sandbox, ["fill", ref, value]);
    }

    await runAgentBrowserCommand(sandbox, ["click", "@e5"]);
    // Pass an app-specific wait, such as ["--url", "**/checkout/complete"],
    // ["--text", "Thanks"], or ["#confirmation"].
    await runAgentBrowserCommand(sandbox, ["wait", ...postSubmitWaitArgs]);

    const ssResult = await runAgentBrowserCommand<{ data?: { path?: string } }>(sandbox, [
      "screenshot",
    ]);
    const ssPath = ssResult.json?.data?.path;
    if (!ssPath) throw new Error("Screenshot did not return a file path.");
    const b64Result = await sandbox.runCommand("base64", ["-w", "0", ssPath]);
    const screenshot = (await b64Result.stdout()).trim();

    await runAgentBrowserCommand(sandbox, ["close"], { json: false });

    return { screenshot };
  });
}

MCP Server over stdin

For many MCP tool calls in one sandbox, startAgentBrowserMcp keeps one agent-browser mcp process running and sends requests over the sandbox command's stdin. It requires @vercel/sandbox 3.7 or later, and timeout (milliseconds) applies to each request and to close():

import { startAgentBrowserMcp, withAgentBrowserSandbox } from "@agent-browser/sandbox/vercel";

const snapshot = await withAgentBrowserSandbox(async (sandbox) => {
  const mcp = await startAgentBrowserMcp(sandbox, { timeout: 60_000 });
  try {
    await mcp.callTool("agent_browser_open", { url: "https://example.com" });
    return await mcp.callTool("agent_browser_snapshot", { interactive: true });
  } finally {
    await mcp.close();
  }
});

Sandbox Snapshots (Fast Startup)

A sandbox snapshot is a saved VM image of a Vercel Sandbox with system dependencies + agent-browser + Chromium already installed. Think of it like a Docker image: instead of installing dependencies from scratch every time, the sandbox boots from the pre-built image.

This is unrelated to agent-browser's accessibility snapshot feature (agent-browser snapshot), which dumps a page's accessibility tree. A sandbox snapshot is a Vercel infrastructure concept for fast VM startup.

Without a sandbox snapshot, each run installs system deps + agent-browser + Chromium (~30s). With one, startup is sub-second.

Creating a sandbox snapshot

The snapshot must include system dependencies (via dnf), agent-browser, and Chromium:

const snapshotId = await createAgentBrowserSnapshot();

Run this once, then set the environment variable:

AGENT_BROWSER_SNAPSHOT_ID=snap_xxxxxxxxxxxx

A helper script is available in the demo app:

npx tsx examples/environments/scripts/create-snapshot.ts

Recommended for any production deployment using the Sandbox pattern.

Authentication

On Vercel deployments, the Sandbox SDK authenticates automatically via OIDC. For local development or explicit control, set:

VERCEL_TOKEN=<personal-access-token>
VERCEL_TEAM_ID=<team-id>
VERCEL_PROJECT_ID=<project-id>

These are spread into Sandbox.create() calls. When absent, the SDK falls back to VERCEL_OIDC_TOKEN (automatic on Vercel).

Scheduled Workflows (Cron)

Combine with Vercel Cron Jobs for recurring browser tasks:

// app/api/cron/route.ts  (or equivalent in your framework)
export async function GET() {
  const result = await withBrowser(async (sandbox) => {
    await sandbox.runCommand("agent-browser", ["open", "https://example.com/pricing"]);
    const snap = await sandbox.runCommand("agent-browser", ["snapshot", "-i", "-c"]);
    await sandbox.runCommand("agent-browser", ["close"]);
    return await snap.stdout();
  });

  // Process results, send alerts, store data...
  return Response.json({ ok: true, snapshot: result });
}
// vercel.json
{ "crons": [{ "path": "/api/cron", "schedule": "0 9 * * *" }] }

Environment Variables

Variable Required Description
AGENT_BROWSER_SNAPSHOT_ID No (but recommended) Pre-built sandbox snapshot ID for sub-second startup (see above)
VERCEL_TOKEN No Vercel personal access token (for local dev; OIDC is automatic on Vercel)
VERCEL_TEAM_ID No Vercel team ID (for local dev)
VERCEL_PROJECT_ID No Vercel project ID (for local dev)

Framework Examples

The pattern works identically across frameworks. The only difference is where you put the server-side code:

Framework Server code location
Next.js Server actions, API routes, route handlers
SvelteKit +page.server.ts, +server.ts
Nuxt server/api/, server/routes/
Remix loader, action functions
Astro .astro frontmatter, API routes

Example

See examples/environments/ in the agent-browser repo for a working app with the Vercel Sandbox pattern, including a sandbox snapshot creation script, streaming progress UI, and rate limiting.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 5aad200. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 1 hour ago.

Activeupdated 3 weeks ago

README badge

README badge for vercel-labs/agent-browser/vercel-sandbox