Smooth CLI Review – Give Your AI Agent a Real Browser

TL;DR

TL;DR: Smooth CLI wraps a specialized browser agent behind a natural language interface, so Claude Code and similar tools can handle web tasks without the token cost and context pollution of low-level click/type automation.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: smooth.sh ← verified 2026-07-23
• Documentation: docs.smooth.sh ← verified 2026-07-23
• CLI docs: docs.smooth.sh/cli/overview ← verified 2026-07-23
• HN launch thread: news.ycombinator.com/item?id=46901233 ← verified 2026-07-23
• Pricing: free plan with welcome credits (verified on smooth.sh homepage)

What Is Smooth CLI?

AI agents like Claude Code are powerful, but they are mostly stuck in the terminal. Meanwhile, most valuable work happens in the browser.

Current browser tools for agents expose low-level actions like click(x=342, y=128), type("search query"), and scroll(down=500). This creates three concrete problems:

| Problem | Why it matters | |---|---| | Slow and expensive | Using a massive model to click buttons wastes tokens and time on every action. | | Context pollution | Every click and keystroke fills the context window with UI noise instead of your actual task. | | Wrong expertise | General-purpose models are not trained to handle iframes, shadow DOMs, and the messy reality of modern websites. |

Smooth CLI flips this around. Your agent says what it wants in plain English, and Smooth figures out how to accomplish it using a specialized model built for browser interaction.

The product scores 92 percent on the WebVoyager benchmark, which tests real-world web navigation tasks. The demo video on the docs page shows an agent autonomously filling out a multi-step form and retrieving a result.

Setup Workflow

Step 1: Get an API Key

Sign up at app.smooth.sh — no credit card required. The free plan includes welcome credits.

Step 2: Install the CLI

npm install -g smooth-cli

Step 3: Configure Claude Code

smooth-cli connect claude-code

This command opens a browser window to authenticate with your Smooth account, then saves the API key to your local Claude Code environment.

Step 4: Run a Task

Once connected, you can prompt your agent to perform web tasks directly:

Search for flights from NYC to LA and find the cheapest option.

The agent sends this natural language instruction to Smooth, which handles the underlying browser steps — clicking, scrolling, filling forms — and returns the result.

Deeper Analysis

Architecture

Instead of giving your agent 200 low-level browser tools, Smooth provides a single natural language interface. A lightweight specialized model handles the clicking so your primary agent can focus on reasoning about goals rather than pixel positions.

Pricing

The free plan includes welcome credits with no credit card required. Paid plans scale by usage. The homepage mentions a $0.02 to $0.04 per task cost range for various automation scenarios.

Alternatives Compared

• —chrome (Claude built-in) — exposes low-level actions, fills context with UI noise.
• Playwright MCP — same problem: hundreds of granular tools that a general-purpose model is not optimized to use.
• browser-use — widely used but significantly slower and more expensive per task.

Smooth claims to be 5x faster and 7x cheaper than browser-use, backed by the WebVoyager benchmark score.

Practical Evaluation Checklist

• [ ] Connect CLI to Claude Code in under 5 minutes
• [ ] Run a research task (e.g., “find the top 5 articles about X this week”)
• [ ] Run a form-filling task (e.g., “get a quote from Octopus Energy”)
• [ ] Monitor live view to verify what the agent sees
• [ ] Check that persistent sessions survive across tasks
• [ ] Verify billing on free credits before paying

Security Notes

Smooth runs browsers in isolated cloud environments with no permissions by default. The P2P Tunnel feature routes traffic through your local machine for geo-restricted content or localhost access — this keeps sensitive credentials off the Smooth infrastructure.

For enterprise use cases, the docs mention security compliance reports, on-prem options, BAA agreements, and SSO.

FAQ

Q: Does Smooth work with agents other than Claude Code? A: Yes. The docs cover OpenClaw and a generic “other agents” setup guide. Any agent that can invoke a subprocess or call an API can use Smooth.

Q: How is this different from Playwright MCP? A: Playwright MCP exposes dozens of low-level actions (click, type, select, hover, scroll, etc.) that a general-purpose model must orchestrate. Smooth exposes one natural language interface. The model describes the goal; Smooth handles the mechanics.

Q: What happens to my browsing data? A: By default, sessions run in isolated cloud environments. The P2P Tunnel option routes traffic through your machine so sensitive sites never touch Smooth infrastructure.

Q: How fast is it compared to browser-use? A: Smooth claims 5x faster and 7x cheaper, citing internal benchmarks. The WebVoyager score of 92 percent is the public benchmark figure.

Conclusion

Smooth CLI solves the browser gap in the AI agent stack by replacing low-level automation primitives with a natural language interface. The token savings and context-window relief are the immediate wins; the reliability improvement (92 percent on WebVoyager) is what makes it production-viable.

If you are running Claude Code or a similar agent today and hitting wall after wall because the agent cannot reliably interact with the web, Smooth CLI is worth a look. The free plan gives you enough credits to validate the workflow before committing to a paid tier.