Implementing Automated Multi-Channel SNS Crawling with Chrome MCP + Claude Code

Introduction

I implemented a system that automatically patrols multiple SNS platforms to collect and record metrics using Claude Code + Chrome MCP (a browser automation tool).

In this article, I will share the technical details of the implementation and the insights gained.

What I Wanted to Do

As an individual developer, I operate the following four platforms:

• BOOTH — Digital product sales
• Zenn — Technical articles
• note — Essays and experiment records
• X (Twitter) — Daily updates

Instead of checking these metrics like sales, likes, and followers manually, I wanted to have an AI agent patrol them automatically.

Architecture

The configuration consists of Claude Code (the "Great Sage") patrolling four platforms via Chrome MCP and recording the retrieved data into registry.yaml.

Patrol Order Design

The patrol order was designed in order of proximity to revenue.

1. BOOTH Management Screen (manage.booth.pm) — Direct revenue
2. Zenn — Technical branding
3. note — Seeds for content revenue
4. X — Customer acquisition channel

Reason: To prioritize data directly linked to revenue and respond early if any issues occur.

registry.yaml — Master Data for Metrics

Data from all platforms is aggregated into a single YAML file. It manages the list of articles, metrics (PVs, likes, follower counts, etc.), and statistical information for each platform.

The reason for choosing YAML over JSON is that it allows comments, is easier for humans to read, and makes git diffs easier to review.

Anomaly Detection Mechanism

Anomalies are detected by comparing with previous values.

• Rapid increase in likes (+50% compared to previous) → High performance notification
• Rapid increase in PVs (+100% compared to previous) → Buzz detection notification

Patrol Results

These are the results of the first full-channel patrol.

• BOOTH: Total sales ¥0 / Items sold 0
• Zenn: 2 articles / 0 likes
• note: 9 articles / 11 total "Suki" (likes)
• X: 12 posts / 0 followers

It is a harsh reality with zero sales and zero X followers, but the biggest achievement is being able to grasp the numbers accurately.

Summary

I implemented a system to automatically patrol four SNS platforms using Chrome MCP + Claude Code.

While technically simple, design decisions such as "patrolling in order of revenue," "centralizing management in YAML," and "systematizing anomaly detection" make operations much easier. As next steps, I plan to implement automatic generation of post drafts and automatic creation of weekly reports.