MCP Gateway Configuration Skill

Context

This skill applies when:

• Configuring multi-server MCP gateway deployments with the EP MCP Server
• Routing MCP tool requests across multiple parliamentary data sources
• Setting up access control policies for different MCP client types
• Integrating the EP MCP Server alongside other MCP servers in a unified gateway
• Configuring tool namespacing to avoid conflicts between MCP servers
• Managing connection pooling and failover for EP API backend connections
• Deploying MCP gateway configurations for development, staging, and production environments

MCP gateways enable AI applications to access multiple MCP servers through a single connection point. The EP MCP Server must be configurable as a backend within larger MCP gateway architectures.

Rules

1. Namespace All Tools: Prefix EP tools with ep_ or europarl_ to avoid naming collisions with other MCP servers in a gateway
2. Define Explicit Tool Routes: Map each MCP tool to its backend server explicitly; avoid wildcard routing that could expose unintended tools
3. Configure Per-Client Access Control: Assign tool access policies based on client identity; not all clients should access all EP data tools
4. Use Environment-Based Configuration: Store gateway configuration in environment variables or secure config files, never hardcoded
5. Implement Health Checks: Configure readiness and liveness probes for the EP MCP Server backend within the gateway
6. Set Timeout Policies: Define per-tool timeout values; EP API calls may have variable latency depending on dataset size
7. Enable Request Correlation: Propagate correlation IDs from gateway to EP MCP Server for end-to-end request tracing
8. Configure Rate Limiting at Gateway Level: Enforce rate limits at the gateway to protect the EP API from excessive downstream requests
9. Support Graceful Degradation: If the EP backend is unavailable, return informative error responses rather than gateway-level failures
10. Version Tool Registrations: Include version metadata in tool listings so clients can detect breaking changes

Examples

✅ Good Pattern: MCP Gateway Configuration

{
  "mcpServers": {
    "european-parliament": {
      "command": "node",
      "args": ["dist/index.js"],
      "env": {
        "EP_API_BASE_URL": "https://data.europarl.europa.eu/api/v2",
        "EP_CACHE_TTL_SECONDS": "900",
        "EP_RATE_LIMIT_PER_MINUTE": "60",
        "NODE_ENV": "production"
      },
      "toolPrefix": "ep",
      "healthCheck": {
        "interval": 30,
        "timeout": 10,
        "path": "/health"
      }
    },
    "national-parliament": {
      "command": "node",
      "args": ["national-server/dist/index.js"],
      "toolPrefix": "nat"
    }
  }
}

✅ Good Pattern: Tool Routing with Access Levels

# Gateway tool routing and access control
tool_routes:
  ep_search_meps:
    backend: european-parliament
    timeout: 15000
    rate_limit: 100/15min
    access: public

  ep_get_voting_records:
    backend: european-parliament
    timeout: 30000
    rate_limit: 50/15min
    access: authenticated

  ep_admin_cache_clear:
    backend: european-parliament
    timeout: 5000
    rate_limit: 5/hour
    access: admin

Anti-Patterns

❌ Bad: Wildcard Tool Exposure

{
  "mcpServers": {
    "european-parliament": {
      "command": "node",
      "args": ["dist/index.js"],
      "exposeAllTools": true
    }
  }
}

❌ Bad: Hardcoded Gateway Configuration

// NEVER hardcode backend URLs or credentials
const EP_SERVER = "http://localhost:3001";
const API_KEY = "sk-hardcoded-key-12345";

❌ Bad: No Timeout Configuration

// NEVER call backends without timeouts
const result = await epServer.callTool(request); // Could hang indefinitely

ISMS Compliance

• AC-003: Per-client access control for MCP tool authorization
• CM-002: Configuration management for gateway deployment environments
• AU-002: Request correlation and audit trail across gateway and backends
• BC-001: Graceful degradation and failover for business continuity

Reference: Hack23 ISMS Policies