Skip to content
← Tools / Skills
ToolSkyvern/10 trustVerified
Skyvern logo

Skyvern

Automate browser workflows with AI — vision LLMs instead of brittle selectors

Visit official site →

About

Skyvern is an open-source AI browser-automation platform that drives real browsers with vision LLMs and computer vision instead of XPath/CSS selectors, so automations survive website layout changes. A swarm of agents (planner–agent–validator) screenshots the page, maps visual elements to actions, and executes through Playwright — it can operate websites it has never seen before.

Verified live 2026-10-13: 23,185 stars, AGPL-3.0 license, YC S23 company, actively maintained (6,981 commits, 159 releases). The headline SDK is a Playwright extension: four AI page commands — page.act(prompt), page.extract(prompt, schema), page.validate(prompt), page.prompt(prompt, schema) — plus page.agent for multi-step tasks, credential-vault login (Skyvern/Bitwarden/1Password, 2FA/TOTP), file downloads, and saved workflows. Every standard Playwright action also accepts an optional prompt parameter (AI element location, or selector-first with AI fallback). Also ships a no-code workflow builder + dashboard, livestreaming, Zapier/Make/n8n integrations, and MCP support. Two ways to run: Skyvern Cloud (managed, API key from app.skyvern.com, bundled anti-bot/proxy/CAPTCHA solvers) or self-host (pip install "skyvern[all]" && skyvern quickstart — SQLite by default, Postgres via flag; or Docker Compose). Bring your own LLM keys when self-hosting.

What it can do

  • Vision-driven automation: screenshots + vision LLMs locate and interact with elements — no XPath/CSS selectors to maintain.
  • Playwright-extension SDK: page.act / page.extract (JSON-schema structured data) / page.validate / page.prompt, plus page.agent.run_task, .login, .download_files, .run_workflow.
  • AI-augmented Playwright: all standard actions accept an optional prompt — pure AI, or selector-first with AI fallback.
  • No-code workflow builder and dashboard for non-technical users; reusable, parameterized workflows.
  • Authentication: credential vault with 2FA/TOTP support and Bitwarden/1Password integrations.
  • Cloud extras: proxy rotation, CAPTCHA solvers, anti-bot/anti-detection mechanisms, parallel instances.
  • Data extraction with schema validation, form filling at scale, file downloads, livestreaming of runs.
  • Integrations: MCP server, Zapier / Make / n8n; Python, TypeScript, and REST APIs.
  • Self-host on your own infra with your own LLM keys for full data privacy.

Setup

  1. Cloud (fastest): pip install skyvern, create an account at app.skyvern.com, and use your API key with the Python SDK, TypeScript client (npm install @skyvern/client), or REST API.
  2. Local UI + server (recommended for self-host): pip install "skyvern[all]" (Python 3.11–3.13; Windows also needs Rust + C++ build tools), then skyvern quickstart — defaults to a local SQLite DB, no Postgres/Docker needed.
  3. Docker Compose (full stack): clone the repo, cp .env.example .env, add your LLM API key to .env, run docker compose up -d, open http://localhost:8080.
  4. MCP route: add the Skyvern MCP server to Claude Code/Desktop or Cursor (streamable-http at https://api.skyvern.com/mcp/ with your x-api-key header).
  5. Optional: set SKYVERN_TELEMETRY=false to opt out of usage statistics.

Prerequisites

Python 3.11, 3.12, or 3.13 (Windows additionally needs Rust and VS Code C++ build tools). An LLM provider API key configured in .env for self-hosted runs. Docker Desktop for the compose path. Note: AGPL-3.0 applies to self-hosted use — running it as a service obligates source sharing. Vision-LLM inference is the main operating cost per task.

Use cases

browser-automationweb-scrapingrpadata-extractionform-filling

Works with

Claude CodeCursorWindsurfCodexChatGPTGeneric APIAny agent (.md)

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: 23,185 GitHub stars (log-scaled vs catalog peers); blended with ~5,856 monthly package downloads; last push 0d ago; 63 releases/12mo; 279 open issues. Manual review: telemetry ON by default (SKYVERN_TELEMETRY=false opt-out); cloud anti-bot/CAPTCHA tooling — ToS/abuse risk; YC S23 verified. License: AGPL-3.0 (copyleft — network/service use triggers source-sharing).

★ 23,185 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.5/10
Maintenance · recency, releases, issue load
9.5/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 23,185 GitHub stars (log-scaled vs catalog peers)
  • · blended with ~5,856 monthly package downloads
  • · last push 0d ago; 63 releases/12mo; 279 open issues
  • · manual review: telemetry ON by default (SKYVERN_TELEMETRY=false opt-out); cloud anti-bot/CAPTCHA tooling — ToS/abuse risk; YC S23 verified

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.