The pattern keeps repeating. Scrapling entered at 92 with 19,000 stars. Kubeshark followed at 84 with 11,800 stars. Now io.github.Skyvern-AI/skyvern arrives with 20,802 stars, 1,845 forks, and a trust score of 79. By star count, it is the largest project to register an MCP server since Scrapling — and by absolute numbers, the most-starred MCP server entry in the registry's history outside the top-25 incumbents.
What Skyvern Is
Skyvern is a browser automation platform that replaces DOM selectors with computer vision. Traditional tools like Selenium and Playwright navigate websites by targeting specific HTML elements — CSS selectors, XPaths, element IDs. When a website redesigns its layout, renames a class, or restructures its DOM, those selectors break. Skyvern takes a fundamentally different approach: it uses visual AI (computer vision combined with LLMs) to see the page the way a human does and interact with it based on visual understanding rather than structural knowledge.
The implications are practical. A Skyvern workflow that fills out an insurance quote form does not break when the form vendor changes its field order or CSS classes. A data extraction task does not fail when the target dashboard gets a UI refresh. The platform handles CAPTCHAs, dynamic content, multi-step workflows, and layout changes that would require constant maintenance with selector-based tools. Key use cases include form filling across varying web interfaces, data extraction from pages that change layouts frequently, and navigating complex multi-step web workflows like procurement portals and insurance quoting systems.
The MCP Integration
With MCP, an AI agent can now delegate browser-based tasks to Skyvern as a tool call. An agent that needs to fill out a web form, extract structured data from a dashboard, or navigate a checkout flow can invoke Skyvern rather than attempting raw browser control. This is a meaningful capability upgrade: most AI agents have no native ability to interact with websites that require authentication, form completion, or multi-step navigation. Skyvern bridges that gap with a visual-first approach that does not require the agent to understand the target site's DOM structure.
The server uses streamable-HTTP transport. No secrets required. 101 watchers. Published under the Skyvern-AI organization namespace.
The Score Breakdown
| Category | Weight | Assessment |
|---|---|---|
| Provenance (30%) | Strong | Source repo present, AGPL-3.0 license, namespace matches org, code of conduct, unique description |
| Maintenance (25%) | Perfect | 52 active commit weeks, 10 contributors, 15 releases/year |
| Popularity (20%) | Very high | 20,802 stars, 1,845 forks, 101 watchers |
| Permissions (25%) | Moderate | Transport risk 10 (streamable-HTTP), credential sensitivity 20, package risk 5 |
| Overall | 79 |
No flags. No secrets. No red flags of any kind.
Why 79, Not Higher
With 20,800 stars and perfect maintenance signals, why does Skyvern land at 79 instead of joining Scrapling in the 80+ tier? Several provenance and metadata factors pull it down. The AGPL-3.0 license is copyleft, not permissive — it requires derivative works to be open-sourced under the same terms, which affects the provenance category score. There is no installable package listed (repo-based install only), no website URL in the registry metadata, no icon, and no SECURITY.md. The permissions category takes deductions for streamable-HTTP transport risk and credential sensitivity. Each of these is a small penalty, but they compound. Scrapling's BSD-3-Clause license, PyPI package, and cleaner metadata profile are why it scored 13 points higher despite having fewer stars.
This is the scoring model working as designed. Stars measure popularity. Trust scores measure the full picture — licensing, packaging, metadata completeness, transport choices, and maintenance signals together. A 20,000-star project with a copyleft license and no package scores lower than a 19,000-star project with a permissive license and clean packaging. The numbers reflect observable facts, not subjective judgment.
The Established-Project Pattern
Skyvern is the fourth established project in three weeks to register an MCP server and land in or near the High Trust tier:
- Scrapling (19.4k stars) — score 92
- Kubeshark (11.8k stars) — score 84
- edgartools (1.8k stars) — score 84
- Skyvern (20.8k stars) — score 79
These are not MCP-native projects. They are proven tools with existing user bases and years of development history that chose to add MCP as an interface. When a project with 20,000 stars and 52 active commit weeks decides the protocol is worth supporting, it is a data point about MCP's trajectory that matters more than any individual score.
Score: 79. No flags. AGPL-3.0.
Sources: Skyvern-AI — GitHub · Scorecard: io.github.Skyvern-AI (score 79)