@ap3x/browser-agent
v0.1.0
Published
Browser-driving agent over @ap3x/browser, with an MCP server
Readme
@ap3x/browser-agent
The agent that drives a browser: a step loop, an action registry, run history, and replay — plus an MCP server exposing it to any MCP client.
Part of AP3X — a TypeScript multi-agent framework.
Install
npm install @ap3x/browser-agentWhat's inside
- Step loop — perceive, decide, act, with a structured output schema and a multi-action executor.
- Action registry —
defineActionwith an explicit context object; actions are filtered per page. - Sensitive data — a placeholder pipeline that keeps secrets out of model context.
- FileSystem workspace — a contained workspace with PDF and DOCX reads.
- Run history + replay — record a run, then replay it deterministically with element re-matching.
- Captcha watchdog — page-state detection that emits a typed signal; it never attempts to solve.
MCP server
npx ap3x-browser-mcpDocumentation
See the AP3X repository for architecture notes, the full package map, and examples.
License
MIT
