Skyvern is not just shipping browser automation. It is trying to make AI agents treat the browser as a callable runtime. The signal is loud: 17 release events in 30 days, 23 GitHub releases in 90 days, 21,647 GitHub stars, a 98 shipping score, and a 97 ToolVitals score.

The interesting part is the shape of the activity. Skyvern’s website positions the product as AI agents that automate workflows on any website, with emphasis on forms, invoices, portal work, data extraction, 2FA, CAPTCHAs, proxy networks, observability, and developer integrations. Its recent blog output narrows that pitch into a specific bet: MCP is becoming the interface layer between AI agents and real browser workflows.

The MCP push is the story

Skyvern published several first-party pieces in May 2026 around MCP architecture and browser automation comparisons. The MCP architecture article says Skyvern’s MCP server exposes 35 tools across browser management, actions, extraction, validation, credential handling, and workflow orchestration. The Browserbase and Puppeteer comparisons repeat the same core argument: selector-driven automation breaks when sites change, while Skyvern wants agents to operate through visual interpretation and goal-level instructions.

That is a sharp product thesis. Skyvern is not competing only with RPA tools or scraping libraries. It is aiming at the messy zone where agents need to log in, click through portals, handle conditional flows, extract structured data, and keep working after the page layout shifts.

The GitHub releases page backs up the sense that the team is still grinding on product mechanics, not just writing positioning pages. The latest visible release, v1.0.36, includes quickstart install work, request policy fixes, LLM override handling, Copilot configuration, classifier evals, API spec updates, and local Chrome remote debugging clarification. That reads like a team tuning the operational edges around agent-driven workflows.

Skyvern is OSI-approved open source under AGPL-3.0, according to the supplied ToolVitals openness data and the repository license page. That matters in this category because browser automation often touches credentials, regulated data, and internal portals. Teams evaluating it can inspect the code path instead of treating the automation engine as a black box.

What ToolVitals cannot infer

ToolVitals can say Skyvern is active, visible, and shipping. It can see release cadence, stars, website availability signals, SSL signals, and the current score model. It cannot prove that Skyvern handles every CAPTCHA, every 2FA flow, or every redesigned portal well in production.

It also cannot measure customer satisfaction, revenue, support quality, hosted uptime beyond tracked availability checks, or whether the visual automation approach is cheaper than maintaining scripts for a given team. The first-party site makes strong claims about CAPTCHAs, 2FA, data extraction, and workflow observability. Those claims are plausible for the product positioning, but they still need workload-specific testing.

The commit count for the last 30 days is not present in the payload, so I would not use Skyvern’s recent activity as a commit-volume argument. The safer read is release and content velocity: 17 release events in 30 days and 23 GitHub releases in 90 days.

The comparison is more nuanced than stars

Against n8n, Skyvern is smaller but more focused. n8n has 188,540 GitHub stars, a 100 shipping score, and 45 release events in 30 days, but ToolVitals classifies it as fair-code, not OSI-approved open source. Skyvern has 21,647 stars, a 98 shipping score, and 17 release events in 30 days, with AGPL-3.0 OSI-approved open source licensing.

Tracecat is a closer openness comparison. It is also AGPL-3.0 OSI-approved open source, with a 98 shipping score and 12 release events in 30 days. Skyvern has far more GitHub stars, 21,647 versus Tracecat’s 3,602, and a slightly higher hot score, 214.7 versus 217.9 for Tracecat means Tracecat edges it on heat while Skyvern has the larger public repo footprint.

Kestra is another useful baseline in automation. It has 26,879 stars, a 100 shipping score, and 12 release events in 30 days under Apache-2.0. Skyvern’s star count is lower, but its recent release-event count is higher. That supports the narrow claim that Skyvern is pushing hard right now, not that it is broadly more mature.

Recommendation

If your team is automating authenticated browser workflows across changing third-party portals, evaluate Skyvern with a real portal task, not a toy scrape. Pick one workflow with login, 2FA or TOTP, file handling, and structured extraction. Run it through Skyvern and through your current Playwright, Puppeteer, RPA, or hosted-browser setup.

Use the result to answer the only question that matters: does Skyvern reduce maintenance when the live web gets weird? The ToolVitals data says the project is active enough to deserve that test. It does not prove the test will pass.

Sources