MCP Server Development

MCP Server Development Skill

🔴 AI FIRST Quality Principle

Apply the AI FIRST principle: never accept first-pass quality. Minimum 2 iterations. Read all output, improve every section. No shortcuts.

Purpose

Guide the development and configuration of Model Context Protocol (MCP) servers for the Riksdagsmonitor platform, enabling AI-powered tooling integration with GitHub Copilot and other MCP-compatible clients.

When to Use

• ✅ Creating new MCP server tools for political data access
• ✅ Configuring .github/copilot-mcp-config.json for new integrations
• ✅ Designing tool schemas for structured data retrieval
• ✅ Implementing MCP transports (stdio, SSE, HTTP)
• ✅ Integrating GitHub MCP tools with CI/CD workflows

Do NOT use for:

• ❌ Standard REST API development (use api-integration skill)
• ❌ UI component development (use vaadin-component-design skill)

MCP Architecture Overview

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  MCP Client     │────▶│  MCP Server  │────▶│  Data Sources   │
│  (Copilot/IDE)  │◀────│  (Tools)     │◀────│  (Riksdag API)  │
└─────────────────┘     └──────────────┘     └─────────────────┘
        │                       │
        │ JSON-RPC 2.0         │ Tool Definitions
        │ over stdio/SSE       │ Input/Output Schemas

MCP Configuration Best Practices

GitHub Copilot MCP Config

{
  "mcpServers": {
    "cia-political-data": {
      "type": "stdio",
      "command": "node",
      "args": ["dist/mcp-server.js"],
      "env": {
        "CIA_DATA_DIR": "${workspaceFolder}/data"
      }
    }
  }
}

Key Configuration Rules

1. Never embed secrets in copilot-mcp-config.json — use environment references
2. Scope tools narrowly — each tool should do one thing well
3. Validate all inputs against JSON Schema before processing
4. Return structured data — prefer typed objects over free-form strings
5. Include error details in tool responses for debugging

MCP Tool Design Patterns

Tool Definition Schema

interface McpToolDefinition {
  name: string;           // e.g., "get_politician_votes"
  description: string;    // Clear, concise purpose
  inputSchema: {
    type: "object";
    properties: Record<string, JsonSchema>;
    required: string[];
  };
}

CIA-Specific Tool Examples

// Politician lookup tool
const getPoliticianTool = {
  name: "get_politician_profile",
  description: "Retrieve profile and voting record for a Swedish Parliament member",
  inputSchema: {
    type: "object",
    properties: {
      personId: { type: "string", description: "Riksdagen person ID" },
      includeVotes: { type: "boolean", default: false }
    },
    required: ["personId"]
  }
};

// Document search tool
const searchDocumentsTool = {
  name: "search_riksdag_documents",
  description: "Search Swedish Parliament documents by keyword, type, and date range",
  inputSchema: {
    type: "object",
    properties: {
      query: { type: "string" },
      docType: { type: "string", enum: ["motion", "proposition", "betankande"] },
      fromDate: { type: "string", format: "date" },
      toDate: { type: "string", format: "date" }
    },
    required: ["query"]
  }
};

Error Handling Pattern

async function handleToolCall(name: string, args: unknown): Promise<McpResult> {
  try {
    const validated = validateInput(name, args);
    const result = await executeQuery(validated);
    return { content: [{ type: "text", text: JSON.stringify(result) }] };
  } catch (error) {
    return {
      content: [{ type: "text", text: `Error: ${error.message}` }],
      isError: true
    };
  }
}

GitHub MCP Integration

Available GitHub MCP Tools

The CIA project uses GitHub MCP server for:

• Repository management — branches, commits, file operations
• Issue tracking — create, update, search issues
• Pull request workflows — reviews, comments, merges
• Actions integration — trigger workflows, check status
• Security scanning — code scanning alerts, Dependabot

Best Practices for GitHub MCP

1. Use specific queries — avoid broad searches that return too many results
2. Paginate results — always handle pagination for large result sets
3. Cache responses — reduce API calls for frequently accessed data
4. Handle rate limits — implement exponential backoff

Security Considerations

• Input validation — sanitize all tool inputs before processing
• Authentication — use token-based auth, never hardcode credentials
• Authorization — scope tool access to minimum required permissions
• Logging — log tool invocations for audit trails, never log sensitive data
• Transport security — use encrypted transports for remote MCP servers

ISMS Alignment