Skip to content
MCP Server Directory logo
automation-scripting

Skyvern

Automate browser based workflows with AI

Reviewed overview

MCP server overview

What is Skyvern? Skyvern is an AI-driven platform to automate browser-based workflows using large language models (LLMs) and computer vision. It provides a Playwright-compatible SDK that adds natural-language and vision-powered interactions on top of Playwright, plus a no-code workflow builder and a managed cloud offering for production/anti-bot needs.

How to use Skyvern?

  • Cloud: sign up at app.skyvern.com to run managed instances with proxies, CAPTCHA solvers, and anti-bot features.
  • Local: install via pip (pip install “skyvern[all]”) or run with Docker Compose. Use the CLI (skyvern quickstart, skyvern run ui/server/all) to launch the server and UI.
  • SDKs: Python package (skyvern) and TypeScript client (@skyvern/client). Typical flow: launch a Skyvern browser, get a page object, then mix Playwright selectors with AI-powered commands to perform tasks.
  • Advanced: connect Skyvern to an existing Chrome via remote debugging or tunnel a local browser to Skyvern Cloud for automating sites with local cookies/logins.

Key features

  • AI-powered Playwright extension: natural-language prompts for actions, extraction, validation, and arbitrary LLM prompts.
  • Core page commands: page.act, page.extract (schema support), page.validate, page.prompt and higher-level page.agent APIs (run_task, login, download_files, run_workflow).
  • AI-augmented Playwright actions: standard actions accept a prompt parameter and support selector-first / AI-fallback modes.
  • No-code workflow builder: chain tasks (browser actions, data extraction, loops, file parsing, HTTP blocks, custom code, uploads) into reusable workflows.
  • Vision-first approach: uses vision LLMs to find and interact with elements, reducing brittleness from XPath/CSS changes.
  • File downloading, data extraction with JSON schema output, livestreaming viewport for debugging, form filling, and robust authentication support (2FA/TOTP, password manager integrations).
  • Integrations: supports many LLM providers, MCP, Zapier/Make.com/N8N, and common password managers (Bitwarden + custom credential service supported).

Use cases

  1. Bulk invoice retrieval and downloading across many provider sites.
  2. Automating job application submission and monitoring.
  3. Automating procurement/materials sourcing workflows in manufacturing.
  4. Filling government forms or registering accounts across changing websites.
  5. Scraping structured data (product info, top posts) and downloading artifacts behind logins.

FAQ (concise)

  • Is Skyvern open source? Yes — the repository is AGPL-3.0 licensed. The managed cloud includes additional anti-bot measures not in the OSS repo.

  • Which LLMs are supported? OpenAI (GPT family), Anthropic (Claude), Azure OpenAI, AWS Bedrock, Gemini, Ollama, OpenRouter, and OpenAI-compatible endpoints via liteLLM/MCP.

  • How do I run it locally? Use pip install “skyvern[all]” and run skyvern quickstart, or use docker compose up -d after cloning the repo and configuring .env.

  • Can Skyvern handle sites behind logins and 2FA? Yes — it supports automated logins, password manager integrations (Bitwarden, custom service), several 2FA flows (TOTP/QR, email, SMS), and can control an existing Chrome containing your session.

  • Is it safe to expose my browser to Skyvern Cloud? You can tunnel a local browser to Skyvern Cloud but must use an API key. The docs warn that a tunnel without auth grants full control; follow security guidance.

  • Can I get structured outputs? Yes — provide a data_extraction_schema (JSON Schema) with tasks and Skyvern will return structured results.

  • How to opt out of telemetry? Set the environment variable SKYVERN_TELEMETRY=false.

For full docs, examples, and troubleshooting see: https://www.skyvern.com/docs and the GitHub repo at https://github.com/Skyvern-AI/skyvern.