Introduction to the Web Agent Protocol (WAP)
‍

We’re excited to introduce WAP (Web Agent Protocol) — an open standard and framework for collecting, interpreting, and replaying real user interactions on the web through autonomous agents. WAP provides the missing bridge between large language models (LLMs) and the real-world web interface — allowing AI agents to learn from human demonstrations and perform tasks reliably in any browser-based environment.

As the capabilities of LLMs continue to improve, one of the remaining frontiers is grounding these models in real, interactive web environments. Many high-value tasks live behind dynamic frontends that are hard to model, brittle to automate, and inconsistent across sessions. WAP solves this with a unified protocol and toolkit that captures action-level intent, DOM context, and environment state, enabling both exact and goal-directed web automation.

The Web Agent Protocol (WAP) is a standardized framework designed to enable seamless interaction between users, web agents, and browsers by recording and replaying browser actions. It separates the concerns of action recording and execution, allowing for efficient automation and reusability. Today, we're introducing two major components of the Web Agent Protocol for developers:

‍

1. A Chrome extension that captures rich, structured interaction data including clicks, text inputs, form submissions, and page content — all labeled and timestamped, with privacy-preserving sanitization.

2. An open-source-repository of WAP servers

Without WAP, training or supervising agents on web tasks involves brittle screen scraping, manual selector engineering, and rigid heuristics. With WAP, every task trace is a structured, interpretable, and replayable artifact. This unlocks a range of applications, such as bootstrapping smart replays that generalize across page changes, and enabling hybrid agents that can combine exact and adaptive strategies

‍

Whether you're developing autonomous browsing agents, building assistants that operate online, or simply experimenting with web-based AI control, WAP gives you the tools to collect data, generate actions, and evaluate performance. You can start a local WAP server, generate exact or smart replay lists from user sessions and run the lists using  agent with additional MCP support.

WAP is open source and designed for community contribution. We’re committed to building a transparent, extensible protocol that serves the research and developer community.

If you're building intelligent web automation, come build with us.  WAP provides a common foundation to build, scale, and share.