Web Agent Protocol

Name: Web Agent Protocol
Author: OTA-Tech-AI

@OTA-Tech-AI

Repository Homepage

About Web Agent Protocol

🌐Web Agent Protocol (WAP) - Record and replay user interactions in the browser with MCP support

#python #mcp #wap #browser-automation #ai-agents #playwright #web-agents #ai-tools #record-replay #llm #mcp-server #modelcontextprotocol #browser-use #web-agent-protocol

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "web-agent-protocol": {
      "command": "python",
      "args": [
        "action_collect_server.py"
      ]
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Web Agent Protocol?

The Web Agent Protocol (WAP) is a standardized framework for recording and replaying browser actions, enabling seamless interaction between users, web agents, and browsers. It separates action recording from execution, providing a Python SDK to collect raw interaction data, convert it into replayable action lists, and optionally export those lists as MCP servers for reuse by any agent.

How to use Web Agent Protocol?

Install dependencies via pip install -r requirements.txt in a Python 3.11 conda environment, set PYTHONPATH to the project root, and create a .env file with your API keys. To record, install the OTA‑WAP Chrome extension, start the data‑collection server (python action_collect_server.py), and browse normally. Generate replay lists with python wap_replay/generate_exact_replay_list.py (exact) or python wap_replay/generate_smart_replay_list.py (smart), then replay with python run_replay.py --model-provider <name> --wap_replay_list <path>. For MCP integration, run python wap_replay/generate_mcp_server.py, then launch python wap_service.py in one terminal and python mcp_client.py in another.

Key features of Web Agent Protocol

Record browser interactions via the OTA‑WAP Chrome extension
Generate exact‑replay or smart‑replay action lists from raw events
Convert recorded actions into MCP servers for reuse by any agent
Replay action lists using the WAP‑Replay protocol
Provide a desktop app for out‑of‑box replay without extra setup

Use cases of Web Agent Protocol

Automate repetitive browser workflows by recording once and replaying any number of times
Create goal‑oriented smart replays to perform complex multi‑step tasks
Convert recorded sessions into MCP servers so agents can reuse them without re‑recording
Collect high‑fidelity user‑interaction data for training or testing web agents

FAQ from Web Agent Protocol

What is the difference between exact replay and smart replay?

Exact replay precisely reproduces every recorded action (clicks, scrolls, etc.), while smart replay condenses the raw event stream into a smaller set of goal‑oriented steps that can adapt to minor page changes.

What are the runtime dependencies for Web Agent Protocol?

The project requires Python 3.11, the packages listed in requirements.txt, and API keys for chosen LLM providers (e.g., OpenAI, DeepSeek) placed in a .env file. A Chrome browser with the OTA‑WAP extension is needed for recording.

How do I use the MCP server conversion feature?

Run python wap_replay/generate_mcp_server.py --task_id <task_id>. The generated MCP server will be placed in the mcp_servers folder. Then start wap_service.py in one terminal and mcp_client.py in another to interact with it via prompts.

Where is recorded data stored?

Each recording session is saved under data/YYYYMMDD/taskid/summary_event_<timestamp>.json. The replay lists are stored in data_processed/exact_replay/ and data_processed/smart_replay/.

I get a "no task‑start file" error when generating replay lists. What does it mean?

The generators require a full recording session that includes exactly one task‑start and one task‑finish event. Ensure the Chrome extension captured a complete session before trying to generate the replay lists.