Smooth CLI – Token-Efficient Browser for AI Agents
Smooth CLI is a serverless browser agent API that lets AI agents like Claude Code navigate the web using natural language instructions, scoring 92% on WebVoyager at 7x lower cost than Browser Use.
TL;DR
TL;DR: Smooth CLI is a browser agent API that lets AI coding agents navigate the web using natural language — scoring 92% on the WebVoyager benchmark at roughly $38 per full benchmark run versus $250 for Browser Use.
Source and Accuracy Notes
⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.
- Project page: smooth.sh — verified via direct fetch
- Docs: docs.smooth.sh — verified via direct fetch
- HN launch thread: news.ycombinator.com/item?id=46901233 — 109 points
- License: not open-source (commercial SaaS API)
- Source last checked: 2026-08-03 (HN story ID 46901233)
What Is Smooth CLI?
Smooth CLI is a serverless browser agent API built by Circlemind Inc that gives AI coding agents like Claude Code a reliable way to navigate the web. Instead of exposing hundreds of low-level browser tools (click, type, scroll, select), Smooth accepts natural language task descriptions and handles the complexity internally.
The core problem it solves: most AI agents are excellent at CLI tasks but helpless when a web-based action is required — filling out a form, scraping behind a login wall, extracting data from a dynamic page. Existing solutions like Playwright MCP or the Vercel agent-browser expose an action space that is too complex for smaller models to use efficiently, and they fail on the “billion edge cases” of the modern web (nested iframes, shadow DOMs, dynamic content).
Smooth takes a different approach: a small, efficient model specifically trained for browsing handles the low-level interaction, keeping the context window clean and the cost low.
Key Features
Token-efficient natural language browsing Agents describe tasks in plain English. Smooth translates these into browser actions without flooding the context window with click-by-click details. According to the HN launch announcement, this approach is “5x faster and 7x cheaper” than using a general-purpose model for browser control.
“Use My IP” for captcha avoidance Smooth can route browser traffic through your own IP address while still running browsers in the cloud. This significantly reduces roadblocks like captchas that plague datacenter-based browser agents.
Session and profile management Browser sessions persist cookies and authentication state. Profiles let the agent maintain logged-in context across tasks, so it can access accounts without re-authenticating on every run.
Custom tools and extensions Register custom browser extensions and API tools that the agent can call during task execution. Files can be uploaded and passed to tasks for processing.
Security by design Running agents on the web means treating them as untrusted actors. Smooth runs in sandboxed virtual browser environments with minimal permissions by default — a security model that aligns with best practices for agentic web access.
Benchmarks
From the product page:
Smooth scores 92% on WebVoyager — the most reliable browser agent benchmark. The cost to run the full WebVoyager benchmark is $38 versus $250 with Browser Use.
This is achieved, according to Smooth’s team, by using small and efficient models purpose-built for browsing rather than relying on large general-purpose models to click through UIs.
Practical Evaluation Checklist
- Does the agent need to interact with a login-gated web page?
- Does the task require navigating complex, dynamic web UIs (React SPAs, shadow DOM sites)?
- Are captchas or IP-based blocking currently blocking your agent workflow?
- Is the agent running up against context window limits when attempting browser tasks?
- Do you need session persistence across multiple browser steps?
If you answered yes to any of these, Smooth CLI is worth evaluating.
Security Notes
Smooth runs browser agents in sandboxed virtual environments. For sensitive tasks, note that zero data retention is available “on request” and is currently limited to the Enterprise plan. The free and Starter plans include community Discord support for troubleshooting.
FAQ
Q: How does Smooth compare to Browser Use by Vercel? A: Browser Use exposes hundreds of low-level browser tools (click, type, scroll, etc.), which bloats the action space and context window. Smooth abstracts all of this into a single natural language task description. On benchmarks, Smooth scores 92% on WebVoyager at roughly 7x lower cost per run.
Q: Does Smooth work with Claude Code? A: Yes. The docs include a dedicated Claude Code integration guide showing how to configure Smooth as the browser backend.
Q: Is it open source? A: No. Smooth is a commercial SaaS product. The core engine is not publicly available.
Q: What does “token-efficient” mean in practice? A: Instead of sending every click, keystroke, and DOM element to the model, Smooth uses a purpose-built small model internally. Only the high-level task description and final results count against your context. This keeps costs and latency dramatically lower than approaches that use large general-purpose models for browser control.
Q: Can I try it before subscribing? A: Yes. The free plan includes welcome credits — no credit card required to get started.
Conclusion
Smooth CLI fills a specific gap in the AI agent toolchain: reliable, affordable web navigation without the context bloat of low-level browser automation. Its 92% WebVoyager score and 7x cost advantage over alternatives make it a practical choice for developers building agents that need to interact with the broader web. The free tier is generous enough for evaluation, and the integration with Claude Code is well-documented.
If your agent workflow is hitting walls because it can’t reliably use a browser, Smooth is worth a look.
TL;DR
TL;DR: Smooth CLI is a browser agent API that gives AI coding agents like Claude Code a token-efficient way to navigate the web using natural language instructions, scoring 92% on WebVoyager at roughly 7x lower cost than alternatives.
Related Posts
ai-setup
Recall – Persistent Memory for Claude Code via MCP Hooks
Recall gives Claude Code a permanent memory store that survives session restarts and context compaction. Four hooks capture and restore context automatically — with cloud SaaS or self-hosted options.
2/28/2026
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026