MCP Instructions Server
Enables teams to share and dynamically update instructions for Claude via a collaborative YAML file, with tools for refreshing and prompts for different modes.
README
MCP Instructions Server
Shared Claude instructions for teams. Edit instructions.yaml, everyone gets updates.
Quick Start
Option 1: Docker (Recommended)
docker compose up -d
# Server runs at http://localhost:8080/mcp
Option 2: Python
# macOS / Linux
./setup.sh
python server.py
# Windows
python -m venv .venv && .venv\Scripts\activate && pip install -r requirements.txt
python server.py
Option 3: Cloud (Railway)
Set environment variables:
MCP_TRANSPORT=streamable-http
PORT=8080
Option 4: Azure Container Instances
./deploy.sh
Requirements: Azure CLI (az) and Docker installed.
The script creates:
- Resource Group
- Azure Container Registry
- Container Instance with public URL
Redeploy after changes:
# Set your ACR name (check Azure portal or use: az acr list -g rg-mcp-server --query "[0].name" -o tsv)
ACR=your-acr-name
# Build, push and restart
az acr login -n $ACR
docker build -t $ACR.azurecr.io/mcp-server:latest .
docker push $ACR.azurecr.io/mcp-server:latest
az container restart -g rg-mcp-server -n mcp-server
Useful commands:
az container logs -g rg-mcp-server -n mcp-server # View logs
az container restart -g rg-mcp-server -n mcp-server # Restart
az group delete -n rg-mcp-server --yes # Delete all
Which Option to Choose?
| Mode | Best For | Cost | Latency |
|---|---|---|---|
| stdio (local Python) | Personal use | Free | ~0ms |
| Docker | Local dev/testing | Free | ~10ms |
| Railway | Quick deploy | Pay per use | ~100ms+ |
| Azure ACI | Team sharing, Azure users | ~$1-2/month | ~100ms+ |
Note: The server only gets called when you invoke a prompt (/team-instructions:production, etc.). Normal Claude messages don't hit the server. Instructions are loaded once into the conversation context.
Connect Claude Code
Create config file:
- Global (all projects):
~/.claude/mcp.json - Per-repo (this project only):
.mcp.jsonin repo root
Remote server (Docker/Railway/Cloud)
{
"mcpServers": {
"team-instructions": {
"type": "streamable-http",
"url": "https://your-server-url/mcp"
}
}
}
Local Python (STDIO)
macOS / Linux:
{
"mcpServers": {
"team-instructions": {
"command": "/absolute/path/to/.venv/bin/python",
"args": ["/absolute/path/to/server.py"]
}
}
}
Windows:
{
"mcpServers": {
"team-instructions": {
"command": "C:\\path\\to\\.venv\\Scripts\\python.exe",
"args": ["C:\\path\\to\\server.py"]
}
}
}
Environment Variables
| Variable | Default | Description |
|---|---|---|
MCP_TRANSPORT |
stdio |
Transport: stdio, streamable-http, or sse |
PORT |
8080 |
Server port (for HTTP transports) |
MCP_PORT |
8080 |
Alternative port variable |
Available Commands
Prompts
| Prompt | Description |
|---|---|
production |
Strict mode with team standards |
freestyle |
No rules, full freedom |
code_review |
Review code with standards |
Tools
| Tool | Description |
|---|---|
help |
List all commands |
refresh |
Reload instructions from file |
Resources
| Resource | Description |
|---|---|
instructions://core |
Core principles |
instructions://production |
Production guidelines |
instructions://checklist |
Delivery checklist |
instructions://all |
All sections combined |
Edit Instructions
- Edit
instructions.yaml - Call
refreshtool (or restart server) - Done - all connected clients get updates
Transports
| Transport | Use Case | Endpoint |
|---|---|---|
streamable-http |
Remote/Cloud (recommended) | /mcp |
sse |
Remote/Cloud (legacy) | /sse |
stdio |
Local Python execution | N/A |
Troubleshooting
| Issue | Fix |
|---|---|
| Server not found | Use absolute paths, check JSON syntax |
| Python errors | Need Python 3.10+, run pip install -r requirements.txt |
| Connection refused | Check MCP_TRANSPORT and port settings |
| Invalid Host header | Server needs MCP_TRANSPORT=streamable-http or sse |
Test Server
# Health check (SSE transport only - not available in streamable-http)
curl http://localhost:8080/health
# Test streamable-http endpoint
curl -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'
Note: The /health endpoint only exists when using SSE transport. For streamable-http, use the MCP endpoint test above.
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。