Guide · Agent skill
How to use the Vercel Sandbox browser automation skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains the vercel-sandbox agent skill, a set of instructions that teaches an AI agent to control headless Chrome inside a temporary Vercel Sandbox virtual machine. It is written for developers who have a Vercel-deployed web app and need server-side browser automation such as screenshots, page checks or form runs. By the end you will know how to install the skill, how to ask for browser work in plain language, and what the skill needs and does on your behalf.
Vercel Sandbox Browser Automationvercel-labs/agent-browser · Apache-2.0 Download the skill Quick start
- 1Download and place the skill. Download the ZIP from this page and unzip SKILL.md into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Install the npm packages. Run pnpm add @agent-browser/sandbox @vercel/sandbox in your Vercel-deployed project.
- 3Configure authentication. Deploy on Vercel to use automatic OIDC, or set VERCEL_TOKEN, VERCEL_TEAM_ID and VERCEL_PROJECT_ID for local development.
- 4Ask for the browser task. Describe the page, the output you want and your framework, using a trigger phrase such as browser automation on Vercel.
- 5Create a snapshot (optional). Run the snapshot helper once and set AGENT_BROWSER_SNAPSHOT_ID for faster startup.
- 6Test and deploy. Run the generated function locally, check the returned title, snapshot or screenshot, then deploy to Vercel.
What the vercel-sandbox skill makes the agent do
The skill teaches an agent how to run agent-browser and headless Chrome inside Vercel Sandbox microVMs. A Linux VM starts on demand, runs the browser commands, and shuts down. The source describes it as working with any Vercel-deployed framework, including Next.js, SvelteKit, Nuxt, Remix and Astro.
- Open a web page and capture its title
- Take a screenshot and return it as base64 data
- Read an accessibility snapshot of a page
- Run multi-step actions such as filling and submitting a form
- Run recurring checks through Vercel Cron Jobs
Install the vercel-sandbox skill
Download the ZIP for this skill from this page. The skill is a single file, SKILL.md. Unzip it into one of these locations: ~/.claude/skills/ for your user account, or .claude/skills/ inside a specific project if you want the skill only for that project.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Per project: unzip into .claude/skills/ in your project folder
- Claude apps: Settings → Capabilities → Skills
Install the project dependencies
The skill's code expects two npm packages and a Node.js and pnpm setup. Run the install command from the source inside your Vercel-deployed project.
- Node.js and pnpm
- npm packages @agent-browser/sandbox and @vercel/sandbox
pnpm add @agent-browser/sandbox @vercel/sandboxSet up authentication
On Vercel deployments, the Sandbox SDK authenticates automatically via OIDC, so you usually need no extra setup. For local development, set VERCEL_TOKEN, VERCEL_TEAM_ID and VERCEL_PROJECT_ID. If these are absent, the SDK falls back to VERCEL_OIDC_TOKEN, which Vercel provides automatically.
- VERCEL_TOKEN: personal access token, for local use
- VERCEL_TEAM_ID: team ID, for local use
- VERCEL_PROJECT_ID: project ID, for local use
VERCEL_TOKEN=<personal-access-token>
VERCEL_TEAM_ID=<team-id>
VERCEL_PROJECT_ID=<project-id>Speed up startup with a sandbox snapshot
Without a sandbox snapshot, each run installs system dependencies, agent-browser and Chromium, which the source says takes about 30 seconds. A sandbox snapshot is a saved VM image with those already installed, and the source says startup then becomes sub-second. Create it once, then set the snapshot ID as an environment variable. The source recommends this for production deployments. The helper script it mentions lives in the demo app.
AGENT_BROWSER_SNAPSHOT_ID=snap_xxxxxxxxxxxxTrigger the skill with natural requests
The agent loads the skill when your request matches its description. The source lists trigger phrases such as "Vercel Sandbox browser", "microVM Chrome", "agent-browser in sandbox" and "browser automation on Vercel". You can also name the task directly, as long as it involves Chrome in a Vercel Sandbox.
Be specific about the page, the output you want and the app framework. That gives the agent what it needs to pick the right pattern.
- Add a screenshot function for https://example.com to my Next.js app using Vercel Sandbox
- Read the accessibility snapshot of my pricing page with agent-browser in sandbox
- Set up a daily Vercel Cron job that checks a pricing page with browser automation on Vercel
Safety notes: scripts and network access
This skill runs scripts and uses the network. The browser runs in an isolated, temporary VM that the source says shuts down after each run, and the skill does not persist browser state between runs unless you use multi-step commands within a single session. Only run browser tasks against sites you have permission to access. The skill can perform real logins and form submissions, which may breach a site's terms of service. Review the generated code before deploying it, especially any code that handles credentials.
What people use it for
- Capture a screenshot of a page in a Next.js app's server action
- Read an accessibility snapshot to check whether a pricing page content has changed
- Fill and submit a form on a site you have permission to use, then capture a confirmation screenshot
- Run a daily Vercel Cron job that reads a page and passes the result to your own alerting code
Frequently asked questions
How do I install the vercel-sandbox agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ inside a project. In the Claude apps, you can upload it under Settings → Capabilities → Skills. The skill is a single file, SKILL.md.
What do I need to use the Vercel Sandbox browser automation skill?
You need Node.js and pnpm, the npm packages @agent-browser/sandbox and @vercel/sandbox, and a Vercel account and project. Local use also needs VERCEL_TOKEN, VERCEL_TEAM_ID and VERCEL_PROJECT_ID. A sandbox snapshot ID is optional but speeds up startup.
Is the vercel-sandbox skill free?
The source material for this skill is licensed under Apache-2.0. The source does not state pricing for Vercel Sandbox or Vercel accounts, so check Vercel's own pricing for usage costs.
Can I use this without a Vercel account?
No. The skill depends on Vercel Sandbox and its credentials, so you need a Vercel account and project. For tasks on your own machine, this skill is not the right fit.
Why does my first run take about 30 seconds?
Without a sandbox snapshot, each run installs system dependencies, agent-browser and Chromium from scratch, which the source says takes about 30 seconds. Create a snapshot once and set AGENT_BROWSER_SNAPSHOT_ID to reduce startup time.
How does this compare to running a browser on my own machine?
This skill runs Chrome in ephemeral Vercel Sandbox VMs that shut down after use, which suits server-side tasks in Vercel apps. It is specific to Vercel infrastructure, so for local browser work or non-Vercel hosting you would need a different approach.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers with a Vercel-hosted app who need server-side browser screenshots, page checks or scheduled monitoring with isolated, ephemeral browsers. It is a poor fit for people without a Vercel account, for browser work on their own machine, or for automating sites they lack permission to access.
Download the skill