ServiceNow MCP Demo — Incident Assistant
An end-to-end Model Context Protocol demo that connects Claude to ServiceNow Incident Management.
README
ServiceNow MCP Demo — Incident Assistant
An end-to-end Model Context Protocol demo that connects Claude to ServiceNow Incident Management. Built to showcase MCP fundamentals (servers, clients, tools, resources, stdio transport, agent loops) applied to a real ITSM workflow.
Architecture
┌─────────────────────────────┐ ┌──────────────────────────┐ ┌────────────────┐
│ Web Client (Streamlit) │ MCP │ MCP Server (FastMCP) │ REST │ ServiceNow │
│ - MCP host + client │◄──────►│ - create_incident │◄──────►│ Table API │
│ - Claude agent loop │ stdio │ - get_incident │ │ (or mock mode)│
│ - Incident cards in UI │JSON-RPC│ - list_recent_incidents │ │ │
│ │ │ - update_incident_state │ │ │
│ │ │ - resource: priority │ │ │
│ │ │ matrix reference │ │ │
└─────────────────────────────┘ └──────────────────────────┘ └────────────────┘
Flow for "my laptop won't boot, open a ticket":
- The web client sends your message to Claude along with the tool schemas it discovered from the MCP server at startup (
tools/list). - Claude decides to call
create_incidentwith inferred urgency/impact/category. - The client forwards the call over MCP (
tools/call); the server hits the ServiceNow Table API (or the in-memory mock). - The result flows back; Claude confirms the incident number; the UI renders an incident card.
Quick start
# 1. Create a virtual environment
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
# 2. Install dependencies
pip install -r requirements.txt
# 3. Configure
cp .env.example .env # then edit: add your ANTHROPIC_API_KEY
# leave SERVICENOW_MODE=mock for a no-setup demo
# 4. Run the web client (it launches the MCP server automatically)
streamlit run client/app.py
Open http://localhost:8501 and try:
- "My laptop won't boot and I have a client demo in an hour — open a ticket."
- "Show me the 5 most recent incidents."
- "Move INC0010001 to In Progress with a work note that the tech is en route."
Live ServiceNow mode
- Request a free Personal Developer Instance (PDI) at https://developer.servicenow.com (takes ~2 minutes).
- In
.env, setSERVICENOW_MODE=liveand fill inSERVICENOW_INSTANCE,SERVICENOW_USER,SERVICENOW_PASSWORD. - Restart the app. Created incidents will appear in your PDI under Incident > All — great for showing the ticket in the real ServiceNow UI during the demo.
PDIs hibernate after inactivity. Wake yours up ~15 minutes before your interview, and keep
mockmode as a fallback.
Testing the server standalone (no UI)
The MCP Inspector is the fastest way to prove the server works on its own:
npx @modelcontextprotocol/inspector python server/servicenow_server.py
You can also add the server to Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"servicenow": {
"command": "python",
"args": ["/absolute/path/to/server/servicenow_server.py"]
}
}
}
Project layout
servicenow-mcp-demo/
├── server/servicenow_server.py # FastMCP server: 4 tools + 1 resource, mock/live backends
├── client/app.py # Streamlit web app: MCP client + Claude agent loop
├── requirements.txt
├── .env.example
├── PROJECT_PLAN.md # Phase-by-phase build plan and status
└── DEMO_SCRIPT.md # 5-minute recorded-demo script
MCP concepts demonstrated
- Server built with FastMCP: tools with typed parameters and docstring-driven schemas
- Resource (
servicenow://reference/priority-matrix) — read-only context, distinct from tools - stdio transport — client launches the server subprocess and speaks JSON-RPC
- Capability discovery — the client never hardcodes tools; it calls
tools/list - Agent loop — multi-round tool use with Claude until a final text answer
- Separation of concerns — the same server works with this web client, Claude Desktop, or MCP Inspector, unchanged
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。