Copilot instructions

1---
2applyTo: "**/mcp/**/*.py,**/mcp/**/*.ts,**/mcp_server/**/*.py,**/.mcp/**"
3---
4 
5# Model Context Protocol (MCP) Development Standards
6 
7## Tool Design
8 
9- Write complete JSON Schema definitions for all tool `inputSchema` — incomplete schemas cause LLM tool misuse
10- Tool descriptions must explain: what the tool does, when to use it, and what it returns
11- Every tool must handle invalid input gracefully — return an error string, not raise an unhandled exception
12- Mark non-idempotent tools (those that cause side effects) explicitly in their description
13- Limit tool scope to the minimum necessary — one tool per logical operation
14 
15## Resource Design
16 
17- Resource URIs must be stable and predictable — avoid timestamp or random-component URIs
18- Always include `mimeType` on resource responses
19- Implement resource listing (`list_resources`) alongside `read_resource`
20 
21## Server Implementation
22 
23- Use official `mcp` SDK (`pip install mcp` / `npm install @modelcontextprotocol/sdk`)
24- Implement graceful shutdown handlers
25- Log all tool invocations with tool name, input hash, and response time (not full input content if PII risk)
26- Version your MCP server — include version in the server name or capabilities
27 
28## Security
29 
30- Validate all file path inputs — never allow path traversal (`../`) in file-reading tools
31- Implement authentication for HTTP-based (SSE) MCP servers
32- Rate limit expensive or external-API-calling tools
33- Never expose credentials in tool schemas, descriptions, or error messages
34 
35## mcp.json Configuration
36 
37```json
38{
39  "mcpServers": {
40    "server-name": {
41      "command": "python",
42      "args": ["-m", "my_mcp_server"],
43      "env": {
44        "API_KEY": "${MY_API_KEY}"
45      }
46    }
47  }
48}
49```
50 
51- Always inject secrets via environment variables — never hardcode in `mcp.json`
52- Pin the server command to a specific executable path or virtual environment in production
53