> Supercharge Playwright with AI commands. Use Skyvern's Python & TypeScript SDKs, REST API, or self-host with Docker. Open source, SOC2 compliant, MCP-ready. [23.0K Stars — See What the Hype's About](https://github.com/Skyvern-AI/skyvern) # The AI browser automation SDK that extends Playwright Drop-in AI commands on top of Playwright. Use natural language to act, extract, and validate — or fall back to selectors. Python & TypeScript SDKs. Self-host or use cloud. Start Building Python & TypeScript Playwright compatible AGPL-3.0 License automate.py 1 3 4 5 7 8 9 ## Playwright, supercharged with AI AI commands added directly to the Playwright page object. Use selectors when you want, natural language when you need, or both as a fallback. Python TypeScript 1from skyvern import Skyvern 2 3skyvern = Skyvern(api\_key\="YOUR\_API\_KEY") 4browser = await skyvern.launch\_cloud\_browser() 5page = await browser.get\_working\_page() 6 7await page.goto("https://example.com") 8await page.act("Log in and navigate to orders") 9data = await page.extract("Get order details", 10 schema\={"order\_id": "string", "total": "number"}) 11is\_done = await page.validate("Order confirmed?") 12await page.agent.run\_task("Download all invoices as PDF") `page.goto(url)` Navigate to any URL "https://example.com" `page.act(prompt)` Perform actions using natural language "Click login and wait for dashboard" `page.extract(prompt, schema)` Extract structured data with optional JSON schema "Get product name and price" `page.validate(prompt)` Check page state, returns boolean "Is the user logged in?" `page.agent.run_task(prompt)` Run a multi-step task end-to-end "Download all invoices as PDF" Traditional Playwright `page.click("#submit")` CSS/XPath selectors — works as expected AI-Powered `page.click(prompt="Submit")` Natural language element location AI Fallback `page.click("#btn", prompt="Submit")` Tries selector first, falls back to AI ## Integrate with everything REST API, native SDKs, MCP for AI agents, and no-code connectors. Use Skyvern however your stack demands. ### REST API Full task and workflow execution via HTTP. Trigger browser automations from any language or platform. `curl -X POST https://api.skyvern.com/v1/tasks` ### Python SDK Native Python client with async support. Extends Playwright with AI commands directly on the page object. `pip install skyvern` ### TypeScript SDK First-class TypeScript client with full type safety. Works in Node.js and serverless environments. `npm install @skyvern/client` ### MCP (Model Context Protocol) Use Skyvern as a tool for any LLM agent. Compatible with Claude, GPT, Gemini, and any MCP-supporting model. `See MCP docs →` ### Zapier / Make / N8N Connect browser automations to 5,000+ apps without writing code. Trigger workflows from any event. `Available on Zapier, Make.com, N8N` ### Workflow Builder Visual no-code builder for chaining tasks. Loops, conditionals, code blocks, file parsing, and more. `Built into Skyvern UI` ## Deploy your way Self-host for full control or use Skyvern Cloud for managed infrastructure. Same powerful SDK, your choice of deployment. Recommended ### Skyvern Cloud Managed · Enterprise-ready \# Connect in one line skyvern = Skyvern(api\_key="your-key") - Managed infrastructure, zero setup - Parallel browser execution at scale - Built-in proxy network & CAPTCHA solving - 2FA and authentication handling - Priority support & SLA - SOC2 Type II compliant Start on Cloud ### Self-Hosted Open Source · AGPL-3.0 \# Get started in 2 commands pip install skyvern skyvern init - Full source code access (AGPL-3.0) - Bring your own LLM (OpenAI, Anthropic, Gemini, Ollama) - Docker Compose or pip install - Your infrastructure, your data - Community support via Discord - Unlimited workflows Clone & Self-Host ## Why developers choose Skyvern A head-to-head look at how Skyvern compares to the tools you've probably already tried. vs.Playwright / Selenium Breaks when selectors change AI locates elements by intent, not DOM Manual maintenance for every UI update Workflows survive redesigns automatically No built-in data extraction page.extract() returns structured JSON vs.Other AI Agents Prompt-only — hallucinate on complex flows Combines vision + code execution + LLM Can't handle login, 2FA, CAPTCHAs Native auth, TOTP, CAPTCHA solving No deterministic replay Cached workflows for repeatable results vs.RPA Tools GUI-based, not version-controllable Code-first, lives in your repo No CI/CD integration pip install / npm install, runs anywhere Vendor lock-in Open source, self-host or cloud ## Technical capabilities Everything you need to build production-grade browser automation — authentication, extraction, debugging, and more. ### Vision-Based Navigation Uses computer vision to understand pages — no DOM coupling or brittle selectors ### Auth & 2FA Support Handles login, TOTP, QR-based 2FA, SMS, and email verification natively ### Workflow Chaining Chain tasks with loops, conditionals, code blocks, HTTP requests, and file parsing ### Structured Extraction Extract data into typed JSON schemas. CSV, tables, nested objects supported ### File Downloading Download files from any website. Auto-uploaded to storage for retrieval via API ### Livestream Debugging Watch the browser viewport in real time. See exactly what the agent sees ### Explainable AI Full chain-of-thought reasoning exposed per step. See exactly why the agent clicked, typed, or navigated — debug like reading a log, not a black box ### Multi-LLM Support OpenAI, Anthropic, Gemini, Ollama, Azure, Bedrock, OpenRouter, and any OpenAI-compatible API ## Built in the open Join thousands of developers building browser automation with Skyvern. Star us on GitHub, ask questions on Discord, or contribute directly. 22,900+ GitHub Stars 5,000+ Discord Members 100+ Contributors AGPL-3.0 Open Source License Star on GitHub Join DiscordRead the Docs ## Frequently asked questions Common questions about Skyvern's SDKs, APIs, and developer tools. ### What SDKs does Skyvern offer? Skyvern provides Python (pip install skyvern) and TypeScript (npm install @skyvern/client) SDKs that extend Playwright with AI-powered commands. Both SDKs add .act(), .extract(), and .validate() methods directly to the Playwright page object. ### Can I self-host Skyvern? Yes. Skyvern is open source under AGPL-3.0 with 20K+ GitHub stars. Deploy with Docker Compose or pip install. Bring your own LLM (OpenAI, Anthropic, Gemini, Ollama). Your infrastructure, your data, with community support via Discord. ### How does Skyvern extend Playwright? Skyvern adds AI commands directly to the Playwright page object. You can use traditional CSS/XPath selectors, AI-powered natural language prompts, or both as a fallback. For example: page.click('#btn', prompt='Submit') tries the selector first, then falls back to AI. ### What is MCP integration? Skyvern supports the Model Context Protocol (MCP), enabling AI assistants like Claude, GPT, and Gemini to use Skyvern as a browser automation tool. This lets LLM-based agents browse, interact with, and extract data from websites. ## Start building in 60 seconds Install Skyvern, launch the quickstart, and run your first AI-powered browser automation. $ pip install skyvern && skyvern quickstart Get Started Try Skyvern Cloud --- - [Sitemap](https://www.skyvern.com/sitemap.xml) - [llms.txt](https://www.skyvern.com/llms.txt) - [Documentation](https://www.skyvern.com/docs)