Smooth - Browser Agent API for AI Tools

TL;DR

TL;DR: Smooth replaces low-level browser automation with a natural-language API, making AI agents 20x faster and 5x cheaper for web tasks.

Source and Accuracy Notes

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

• Project page: smooth.sh
• Source repository: github.com/circlemind-ai/smooth-sdk
• License: Unknown (not specified in README or repository metadata; verify before citing)
• HN launch thread: news.ycombinator.com/item?id=46901233 (109 points)
• Source last checked: 2026-08-10 (commit verified against GitHub API)

What Is Smooth?

Smooth is an AI browser automation SDK and CLI published by Circlemind. Rather than exposing low-level browser primitives like clicks, scrolls, and form fills, it accepts natural-language goals and handles the DOM complexity internally.

The core problem it solves is a mismatch in abstraction level: existing browser tools expose hundreds of micro-actions (click at x/y, type into field, scroll N pixels), forcing a large language model to orchestrate mechanics instead of goals. Smooth flips this by letting you write:

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

And handling the underlying browser steps automatically.

Setup Workflow

Step 1: Install the CLI

pip install smooth-py

Step 2: Configure your API key

smooth config --api-key <your-key>

Get your key at app.smooth.sh.

Step 3: Add as a Claude Code skill

npx skills add https://github.com/circlemind-ai/smooth-sdk

This integrates Smooth with Claude Code, allowing it to delegate web navigation tasks.

Step 4: Run a session

# Start a session
smooth start-session

# Run a task
smooth run <session-id> "Find the pricing page and extract all plan names and prices" --url "https://example.com"

# Close when done
smooth close-session <session-id>

How It Works

Traditional browser automation via MCP or --chrome flags produces chains like:

click(x=342, y=128)
type("search query")
click(x=401, y=130)
scroll(down=500)
click(x=220, y=340)

Each step pollutes the model context window and requires the model to reason about low-level mechanics. Smooth abstracts this into goal-oriented calls, keeping the model focused on what it needs rather than how to achieve it.

From the README:

Today, Smooth is 20x faster and 5x cheaper than Claude Code with --chrome.

The CLI can also use your IP address to avoid captchas, and supports concurrent browser sessions for parallelized tasks.

Python SDK

Beyond the CLI, Smooth offers a Python SDK for programmatic use:

from smooth import SmoothClient

smooth_client = SmoothClient(api_key="cmzr-YOUR_API_KEY")
task = smooth_client.run("Go to google flights and find the cheapest flight from London to Paris today")

print(f"Live URL: {task.live_url()}")
print(f"Agent response: {task.result()}")

The SDK supports async clients, persistent sessions, auto-captcha solving, stealth mode, and horizontal scaling.

Supported AI Tools

The README lists compatibility with:

• Claude Code
• Clawdbot / Moltbot / OpenClaw
• Codex
• Cursor
• Antigravity
• Cline
• Factory AI
• GitHub Copilot
• Kiro
• OpenCode
• Windsurf
• Any agent that can run CLI commands

Practical Evaluation Checklist

• Native language goal interface vs. DOM primitives
• Token efficiency compared to Playwright MCP
• Captcha avoidance in practice
• Concurrent session performance
• Stealth mode effectiveness

Security Notes

• API key stored via smooth config or CIRCLEMIND_API_KEY environment variable
• No mention of data retention or session logging in the README
• Review app.smooth.sh privacy policy for cloud sessions

FAQ

Q: Is Smooth open source? A: The CLI and Python SDK are published on GitHub under the circlemind-ai/smooth-sdk repository. The underlying browser agent may be a hosted service (the README references an API key from app.smooth.sh). The license field was not populated in the repository metadata.

Q: How does it compare to Playwright MCP? A: Playwright MCP exposes low-level browser actions (click, type, select) that require a model to orchestrate mechanics. Smooth abstracts this into goal-oriented calls, reducing context overhead per task.

Q: Does it work with existing CI pipelines? A: Smooth is primarily designed for AI agent integration rather than traditional CI. For self-hosted CI alternatives, see Cicada (FOSS GitHub Actions alternative) on HN.

Q: What is the pricing model? A: Requires an API key from app.smooth.sh. The README does not specify self-hosted or open-source deployment options for the underlying agent engine.

Conclusion

Smooth addresses the abstraction mismatch in AI browser automation by providing a goal-oriented interface that keeps large language models focused on tasks rather than mechanics. The 20x speed and 5x cost claims relative to Claude Code --chrome are notable, though verify these against your own workload. The SDK and CLI are available now on GitHub.

For a full technical deep-dive and performance benchmarks, see the official performance summary.