perzianhq/mantis is a browser interface for AI agents. It converts browser pages into numbered text so that agents read and manipulate the content instead of taking screenshots of the screen. The latest release is v0.1.13, but the provided material contains no change history for this version, so nothing can be confirmed about what was newly added or modified.
The tool offers a browser automation approach that handles page content as text. It ships both an MCP server and direct read commands, and the README claims navigation is 36% faster than Playwright MCP. Comparison results against accessibility-tree-based tools are also published, giving material for gauging where this approach sits. It is a TypeScript project using the MIT license. The documentation also covers the browser operation flow, a security threat model, and instructions for installation and self-testing.
However, the documentation also states the limits of the testing: it has constraints on canvas manipulation and on handling logins and bot-blocking. The 36% is therefore a comparison result presented by the README, not a performance figure verified across all environments. From this material alone it is hard to judge the scope of validity of the comparison figure or the specific changes in v0.1.13. The scope of the text-based approach is clear, but whether it can take on complex browser tasks is not confirmed by the published material.