Pilot Browse MCP

中文

MCP toolset + website capability learner for the browser. Agent autonomously explores websites and builds understanding manuals.

Showcase (Pi Demo)

Capability Learning

Install a skill. As the agent interacts with a website, it automatically discovers APIs and binds them with browser workflows into reusable capabilities.

Learning output is saved in website-manuals/:

website-manuals/<site>/
  README.md              # Root index
  pages/                 # Page interaction models
  navigation/            # Navigation paths
  workflows/
    README.md            # Workflow index
    flows/               # Workflow JSON files
    scripts/             # PAB automation scripts (.pab files)
  apis/
    README.md            # API index (browse first)
    endpoints/           # API JSON files

Guided Teaching

(Using Bilibili as demo site)

Stuck on complex interactions? Record or mark elements manually to help the agent learn.

Mark Element

Record Workflow

Manual Reuse

Based on the exploration manual, the agent can quickly understand websites and complete tasks faster.

Two no-LLM modes are also supported:

• .pab scripts for automation tasks -- capabilities match the agent's MCP tools
• Python/other scripts for specialized data collection tasks

Demos

Build various workflows, simple demo showcase.

Automation Operations

Agent: Qidian Novel Saver

Search a novel, save the first 5 chapters.

.pab Script: Bilibili Open 3 Trending Videos and Comment

Python Script Generation

Generate various scripts based on the exploration manual.

For example, a novel scraper script:

I used Pilot to analyze the novel website structure, chapter API, generated scraping scripts, built Docker images, deployed 9 containers on NAS, and crawled over 4000 novels in 2 days.

Install

# Build
cd server && npm install && npm run build
cd extension && npm install && npm run build

# Load extension
# chrome://extensions/ -> Developer mode -> Load unpacked -> extension/dist/

# Start server
cd server && node dist/index.js

MCP Configuration

Example - fill args with the actual path to server/dist/index.js:

{
  "mcpServers": {
    "browser-mcp": {
      "command": "node",
      "args": ["/path/to/server/dist/index.js"]
    }
  }
}

Skills

Check scripts/Skill for available skills.

Quick Start with Examples

Refer to agent-examples/ for ready-to-use agent workspace examples. Run your agent in one of those directories and it will automatically load the MCP config, skills, and project prompts. (Remember to update the MCP path to your actual setup.)

Architecture

AI Agent (Claude Code / Pi / Codex)
    |
    | 1) MCP stdio protocol (JSON-RPC)
    | stdin / stdout
    v
MCP Server (Node.js)          protocol translator
    |
    | 2) WebSocket :9456
    v
Chrome Extension
    |
    | 3) Chrome API
    |
    v
Browser

Features

• 62 MCP tools -- tab management, content extraction, DOM operations, network interception, file saving, workflow recording, script automation, SQL injection detection, JS reverse
• PAB scripting -- Python-like DSL for browser automation with if/for/fn support. Run from popup, no LLM needed
• Element picker -- click any element on the page and tell the agent what it is
• Workflow recording -- demonstrate operations to the agent, it learns and reuses
• Network API toolkit -- monitor, search, inspect, replay with overrides, export code, and analyze site API structure
• Token-efficient saving -- save page content directly to disk, bypassing the LLM
• Shadow DOM + contenteditable

Tools

License

MIT