Enterprise WeChat Document Read-Write Assistant

The Problem to Solve

In many enterprise workflows, core documentation is stored in Enterprise WeChat's online documents (doc.weixin.qq.com). When developers aim to build AI Agents that automate interactions with these documents—such as extracting tasks from meeting minutes, appending records to data tables, or generating periodic report screenshots—they encounter several practical difficulties: Enterprise WeChat documents lack an open API, login verification relies on WeChat QR code scanning, and different document types (docs, sheets, slides, mindmaps) have varying structures and interaction patterns. Manually operating through a browser cannot meet automation needs, while simulating login and parsing page elements requires handling complex session states and page elements.

Core Capabilities and How It Works

The Enterprise WeChat Document Read-Write Assistant solves these issues through an MCP (Model Context Protocol) service. It encapsulates browser automation (based on Playwright) into a set of clear tools for AI Agents to call. Its core workflow revolves around three key steps:

• Automated Login Management: The Agent first calls wecom_doc_status to check the current session. If not logged in or the session has expired (returning "未登录或登录已过期"), it automatically triggers wecom_doc_login, launching a browser window for the user to scan the QR code. Upon successful login, the Cookie is saved locally for several days, eliminating the need for repeated logins. This process is fully handled in the background; the user only needs to scan the code.
• Document Content Reading: For any given Enterprise WeChat document link, the Agent uses the wecom_doc_fetch tool to scrape content. For online docs (doc type), it returns plain text. For online sheets (sheet type), it returns structured information including a list of all worksheets and cell data, with support for specifying a particular worksheet via the tab parameter.
• Structured Content Writing: Write capabilities are specifically designed based on document type. For online docs, wecom_doc_write allows appending plain text to the end of the document. For online sheets, wecom_doc_write_sheet provides more granular operations: you can specify a cell coordinate (e.g., A1, N5) and a write mode (append to add or overwrite to replace). This enables batch row insertion—first reading to confirm the last row number, then writing new data cell by cell.

Additionally, the wecom_doc_screenshot tool can capture page screenshots, useful for visual data capture scenarios.

Use Cases and Boundaries

This skill is highly suitable for scenarios requiring integration of Enterprise WeChat documents into AI workflows, such as:
- Periodically fetching data from daily reports to other systems.
- Automatically appending key points from meeting summaries to a team-shared meeting minutes document at the end.
- Performing automated data updates and archiving on critical business spreadsheets.

However, it's important to understand its limitations and boundaries:
- Limited Write Capabilities: Online docs (doc) only support appending text at the end; they do not support inserting or modifying content at arbitrary positions. Writing to online sheets (sheet) operates cell by cell, which may be inefficient for large-scale data writes.
- Read-Only Types: Slides (slide), mindmaps (mindmap), and flowcharts (flowchart) currently support reading only (e.g., extracting text, nodes, or screenshots) and do not support content writing.
- Browser Environment Dependency: The entire service is built on Playwright and Chromium browser, so initial installation requires downloading browser binaries, and runtime requires corresponding system resources. Login relies on WeChat QR code scanning and cannot be fully automated in the background.
- Session Validity Period: Cookies remain valid for about a few days; once expired, QR code re-login is needed, but the Agent handles this automatically.

Before use, ensure your AI tool (e.g., CodeBuddy, Cursor) has the MCP service properly configured, and understand the above capability boundaries to avoid expectation mismatches.