MCP Server
Skyvern-AI/skyvern
Skyvern-AI/skyvern· Python
LLM + vision browser automation that survives DOM changes, with an MCP server for agent use.
GitHub stats
- Stars
- 21,428
- Forks
- 1,969
- Watchers
- 104
- Open issues
- 154
meta
- License
- AGPL-3.0
- Primary language
- Python
- Last commit
- 2026-04-29
- Stats fetched at
- 2026-04-29
Skyvern automates browser workflows using LLMs and computer vision instead of brittle XPath/CSS selectors, so flows keep working when sites change. It ships a Python/Playwright engine, a workflow builder, REST API, and an MCP server that lets Claude Code, Cursor, or any MCP client drive a real browser. Self-host via Docker or use Skyvern Cloud; reports 85.85% on WebVoyager.
Editor's verdict
Pick Skyvern when you need long-running, repeatable workflows on sites you don't control — form filling, portal scraping, RPA-style tasks — and want them to survive UI churn. Trade-off: vision calls cost real tokens and latency, and AGPL-3.0 means you'll want a commercial license for closed SaaS embedding. For one-off "let an agent click around" use cases, Browser Use or Playwright MCP are lighter; for pure HTML scraping, Firecrawl is cheaper.