Student Management MCP Prototype

Objective

For the "student growth profile and proactive support closed loop" scenario, provide a set of MCP tools already integrated with the agent platform, enabling agents to query student profiles, filter watchlists, create and track support tasks, write support records, generate class dashboards, and integrate with DingTalk groups.

Related MCP server: TeacherAssist-MCP

Highlights

• Upgrade from "Q&A-style student management" to "data-driven proactive discovery."

• Combine attendance, grades, dormitory feedback, practical training performance, financial aid status, and heart-to-heart talk records into a student growth profile.

• Support generating a "today's watchlist," reflecting the shift in student management from passive response to proactive alerting.

• Support writing support records, forming a closed loop of "identify issues - make recommendations - implement follow-up - backfill records."

• Can generate class dashboard data, suitable for competition defense presentations.

Tool List

Local Run

cd D:\教育信息技术应用大赛\学生管理MCP原型
python -m uvicorn student_management_mcp_api:app --host 127.0.0.1 --port 8765

The public deployment and the agent platform use student_management_mcp_sse:app; for local verification you can run:

python -m uvicorn student_management_mcp_sse:app --host 127.0.0.1 --port 8000

Data Persistence

• When DATABASE_URL is configured, PostgreSQL is used, and source ledger ingestion, tasks, follow-up records, and audit ledgers are preserved across Render deployments.

• When not configured, the local SQLite .data/student_management.db is used automatically, which is suitable for development testing, but data may be lost after a Render redeployment.

• On first database initialization, the base data in followup_records.json and support_tasks.json in the repository is automatically migrated.

• The first deployment also reads the 5 Excel files, 1 Word file, and 1 PDF in 演示业务源文件, normalizes the competition demo rosters, attendance/academics, practical training, screening recommendations, talks, tasks, and training room resource scheduling ledgers, and writes them to the database; talks and tasks subsequently added via MCP continue to accumulate.

• The /health health check returns the storage type and record counts, but does not return the database address or DingTalk secrets.

Example PostgreSQL connection variables:

DATABASE_URL=postgresql://用户名:密码@主机:5432/数据库名

Demo Data Maintenance

All content in 演示业务源文件 is simulated data for competition demonstration and does not correspond to real individuals. The recommended screen-recording flow is:

工作人员维护 Excel / Word / PDF 台账
  -> 服务自动解析并入库 PostgreSQL
  -> 班级态势与学生成长时间线
  -> 生成帮扶任务并回写谈话记录
  -> 按期复访提醒与任务审计
  -> 按日期、时段、容量和设备条件查询可用实训室

You can first call get_data_ingestion_status to show files, ingested counts, and the data pipeline, then call the controlled query tools by role:

• get_authorized_student_growth_timeline: students are limited to themselves; class teachers and counselors are limited to authorized classes.

• get_authorized_class_operational_records: class teachers and counselors view activity in authorized classes; administrative staff see only anonymized aggregates.

• list_authorized_followup_reminders: class teachers and counselors view revisit to-dos within their authorized scope.

Restore baseline data:

python demo_data_maintenance.py --mode baseline --confirm RESET-DEMO-DATA

Generate two fixed demo tasks:

python demo_data_maintenance.py --mode sample --confirm RESET-DEMO-DATA

This script clears the current tasks and the audit ledger, so it may only be used by maintenance personnel during screen-recording preparation; it is not an MCP tool and does not send DingTalk messages.

Call Examples

Query student profile:

{
  "tool": "get_student_profile",
  "arguments": {
    "query": "S004"
  }
}

Query the watchlist:

{
  "tool": "list_attention_students",
  "arguments": {}
}

Write support record:

{
  "tool": "create_followup_record",
  "arguments": {
    "student_id": "S002",
    "owner": "辅导员",
    "summary": "已完成首次谈心，学生反馈近期课程压力较大，实训日志补交意识不足。",
    "next_action": "联系实训指导教师确认补交要求，三天后复访。",
    "status": "跟进中"
  }
}

Generate class dashboard:

{
  "tool": "get_class_dashboard",
  "arguments": {
    "class_name": "智能制造2401"
  }
}

Create support task (no push by default):

{
  "tool": "create_student_support_task",
  "arguments": {
    "student_query": "S004",
    "owner": "辅导员",
    "due_date": "2026-07-18",
    "push_to_dingtalk": false
  }
}

Complete the task and sync the support record:

{
  "tool": "update_student_support_task",
  "arguments": {
    "task_id": "任务编号",
    "status": "已完成",
    "progress_note": "已完成线下谈心。",
    "next_action": "联系任课教师并于一周后复访。",
    "sync_followup": true
  }
}

Platform Integration Suggestions

The platform has been verified to support two entry points:

1. Manually create an MCP service

   • Service name

   • Description

   • Connection method: sse

   • API endpoint

   • Header configuration

   • Click test link

2. Import from JSON

   • The format is mcpServers

   • See mcp_servers.example.json for an example

The current SSE MCP service is deployed to Render, and the platform discovers and calls tools via the /sse address. Local 127.0.0.1 is only for development verification and cannot be used as the platform's public access address.