How AI + Enterprise WeChat Cut Procurement Monitoring from Days to Hours

Problem

Manual tender monitoring requires scanning dozens of portals (national procurement site, provincial resource‑trading centers, industry platforms). This creates three core issues: scattered information makes manual retrieval inefficient, deadlines are missed when opportunities disappear quickly, and compiling reports for leadership is labor‑intensive.

Architecture

Two‑layer design separates concerns:

Bottom layer – Dify tender‑monitoring workflow connects to Alibaba Cloud Waterdrop Credit MCP service, which provides over 150 million structured tender records via a single API, eliminating crawling and pagination.

Top layer – Enterprise WeChat bot pushes daily reports to a group chat and handles on‑demand queries via @‑bot, requiring no additional app or web page.

Enterprise WeChat user → WSS long‑connection → Node.js relay → Dify workflow → MCP tender database

Dify workflow

The workflow consists of six nodes:

Start → Date Calculation → MCP Search Tender → Filter & Format → Daily Report → End

Input parameters include keywords, date range, region, and budget.

The key node calls the MCP search_bid API, which returns a single structured response containing title, purchaser, budget amount, deadline, and source link.

Filter & Format node (Python) performs:

Region and budget filtering

Keyword relevance sorting (high / medium / low)

Exclusion of already awarded notices

Automatic conversion of budget units

Why MCP over web crawling

Initial attempt with a SearXNG search‑engine pipeline encountered multiple failures:

Frequent CAPTCHA on Baidu caused scripts to stop.

Search results mixed official and unofficial sources, requiring manual verification.

No budget data in the primary results, necessitating a second‑level page scrape.

The overall pipeline contained more than a dozen fragile nodes.

Switching to Alibaba Cloud MCP yielded:

One API call returns fully structured data.

Workflow reduced from dozens of nodes to 6 nodes.

Stability and data quality improved dramatically.

Enterprise WeChat bot implementation

The relay runs in a single server.js file using only native Node.js modules—zero npm dependencies. Execution command: node server.js A WebSocket (WSS) long‑connection is used instead of traditional HTTP callbacks. Advantages:

Traditional HTTP callback requires a public IP; WSS initiates the connection to the WeChat server, so no public IP is needed.

HTTP callbacks need explicit message encryption/decryption; WSS is encrypted by default.

Configuration complexity drops to only BOT_ID and BOT_SECRET.

In‑chat keyword management

Keywords are stored in a JSON file and can be managed directly in the chat without restarting the service. Example interaction:

👤 查看关键词 🤖 📌 当前每日推送关键词 (3个) 1. 信息化建设 2. 文明城市创建 3. 智慧城市 👤 添加 数字化转型 🤖 ✅ 已添加关键词：数字化转型 👤 删除 文明城市创建 🤖 ✅ 已删除关键词：文明城市创建

Natural‑language commands such as “查看关键词”, “添加 数字化转型”, and “删除 文明城市创建” are recognized, and changes take effect immediately.

Deployment (5 minutes, 4 steps)

Clone the repository to the target server.

Fill in four configuration items: BOT_ID, BOT_SECRET, DIFY_API_KEY, WEBHOOK_URL.

Install the script as a systemd service.

Verify the health‑check endpoint to confirm the service is running.

The service starts on boot and automatically restarts on failure, providing zero‑maintenance operation.

Operational impact

Every morning at 8:30 am the group receives an automated tender report with direct source links. On‑demand queries are handled in seconds; for example:

查询 智慧交通 江苏 30

returns a complete report within a few seconds.

Measured benefits:

⏱️ Saves 1–2 hours of manual information retrieval per day.

🎯 Critical projects are no longer missed; response time shrinks from "days" to "hours".

📊 Leadership views aggregated results directly in the chat, eliminating separate report generation.

Extensibility

The same two‑layer architecture can be repurposed for other automation scenarios, such as:

Public‑opinion monitoring bot (brand sentiment, industry news).

Policy‑update bot (government subsidies, filing notices).

Competitor‑tracking bot (award notices, product releases).

Daily industry‑report aggregation.