web agent protocol

Report Abuse

Basic Information

The Web Agent Protocol (WAP) repository provides a standardized framework and a Python SDK to record, convert, and replay user interactions in web browsers. It is designed to separate recording from execution so captured browser events can be transformed into reusable action lists and served to agents or users. The repo includes a Chrome extension for capturing interaction data, a data collection server to receive event streams, generators to produce either exact-replay or smart goal-oriented replay lists, utilities to convert recordings into MCP servers, and a replay engine that executes recorded actions against a browser. The README documents setup steps, environment configuration, example commands for generating replays and MCP servers, file output conventions for recorded sessions, and troubleshooting tips for running the tooling on a developer machine.

Links

App Details

Features
The project bundles components for end-to-end browser action capture and replay. It provides a Chrome extension to capture DOM events, a Python action data collection server listening on a local endpoint for session files, and a standardized JSON event format saved per task. It includes scripts to generate two replay modes: exact replay that reproduces every recorded action and smart replay that condenses steps into goal-oriented actions. There are tooling scripts to convert replay lists into MCP servers and to run replay via a WAP replay protocol. Additional conveniences include example commands for running replay and MCP clients, a desktop replay tool release for non-developers, organized output folders for processed data, and basic troubleshooting guidance.
Use Cases
WAP helps developers and researchers build reproducible browser automation and agent workflows by turning human interactions into machine-executable action lists. Recorded sessions can be replayed exactly for debugging or transformed into higher-level smart steps for agent-driven task execution, enabling reuse across agents and users. Converting recordings into MCP servers allows model contexts to serve standardized browser operation endpoints. The toolkit reduces the engineering effort to capture realistic interaction traces, test agent behaviors against real sessions, and run controlled replays locally or via the provided desktop tool. Documentation on environment setup, example commands, and output structures assists teams in adopting the protocol for data collection, testing, and automation.

Please fill the required fields*