mcp-standards

# MCP Standards Skill

## 1. Overview

### What is MCP in Claude Code?

Model Context Protocol (MCP) is the standard way to extend Claude Code with custom tools and integrations. MCP servers provide:

- **Tool Integration**: Connect to external APIs, databases, and services
- **Context Providers**: Supply relevant information to Claude during conversations
- **Action Handlers**: Execute operations in external systems
- **Data Sources**: Access project-specific or organization-specific data

### Why Standardization Matters

Standardized MCP servers ensure:

1. **Predictable Behavior**: Developers know what to expect from MCP tools
2. **Easier Debugging**: Consistent patterns make issues easier to identify
3. **Better Discoverability**: Standard naming helps Claude and users find tools
4. **Maintainability**: Common patterns reduce maintenance burden
5. **Team Consistency**: Multiple developers follow same conventions

### MCP in the Plugin Ecosystem

MCP servers are plugin components alongside agents, commands, and skills:

```
plugin/
├── agents/          # Specialized Claude instances
├── commands/        # CLI commands
├── skills/          # Knowledge documents
└── mcp-servers/     # MCP tool providers ← We're here
```

**Key Difference**: While agents use built-in tools, MCP servers provide NEW tools that extend Claude's capabilities.

---

## 2. MCP Server Structure

### Standard Directory Layout

```
mcp-servers/
├── server-name/
│   ├── index.ts                  # Server entry point
│   ├── package.json              # Dependencies and metadata
│   ├── tsconfig.json             # TypeScript configuration
│   ├── README.md                 # Server documentation
│   ├── tools/                    # Tool implementations
│   │   ├── read-tool.ts
│   │   ├── write-tool.ts
│   │   └── index.ts              # Tool exports
│   ├── lib/                      # Shared utilities
│   │   ├── client.ts             # API client
│   │   ├── validation.ts         # Input validation
│   │   └── errors.ts             # Error handling
│   └── tests/                    # Test files
│       ├── read-tool.test.ts
│       └── write-tool.test.ts
```

### Entry Point Pattern (index.ts)

```typescript
#!/usr/bin/env node
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
  CallToolRequestSchema,
  ListToolsRequestSchema,
} from "@modelcontextprotocol/sdk/types.js";

// Import tools
import { fetchTool, createTool, updateTool } from "./tools/index.js";

const server = new Server(
  {
    name: "mcp-plugin-server",
    version: "1.0.0",
  },
  {
    capabilities: {
      tools: {},
    },
  }
);

// Register tools
server.setRequestHandler(ListToolsRequestSchema, async () => ({
  tools: [fetchTool.definition, createTool.definition, updateTool.definition],
}));

server.setRequestHandler(CallToolRequestSchema, async (request) => {
  const { name, arguments: args } = request.params;

  switch (name) {
    case fetchTool.name:
      return fetchTool.handler(args);
    case createTool.name:
      return createTool.handler(args);
    case updateTool.name:
      return updateTool.handler(args);
    default:
      throw new Error(`Unknown tool: ${name}`);
  }
});

// Start server
async function main() {
  const transport = new StdioServerTransport();
  await server.connect(transport);
}

main().catch(console.error);
```

### Tool Module Pattern

```typescript
// tools/fetch-tool.ts
import { z } from "zod";

const inputSchema = z.object({
  id: z.string().describe("The resource ID to fetch"),
  includeMetadata: z.boolean().optional().describe("Include metadata in response"),
});

export const fetchTool = {
  name: "mcp__plugin__fetch_resource",
  definition: {
    name: "mcp__plugin__fetch_resource",
    description: "Fetch a resource by ID from the external service",
    inputSchema: {
      type: "object",
      properties: {
        id: {
          type: "string",
          description: "The resource ID to fetch",
        },
        includeMetadata: {
          type: "boolean",
          description: "Include metadata in response",
        },
      },
      required: ["id"],
    },
  },
  handler: async (args: unknown) => {
    const validated = inputSchema.parse(args);

    try {
      const result = await fetchResourceById(validated.id);
      return {
        content: [
          {
            type: "text",
            text: JSON.stringify(result, null, 2),
          },
        ],
      };
    } catch (error) {
      throw new Error(`Failed to fetch resource: ${error.message}`);
    }
  },
};
```

---

## 3. Tool Naming Conventions

### Standard Pattern

```
mcp__<plugin-name>__<tool-name>
```

**Components**:
- `mcp__` - Universal prefix indicating MCP tool
- `<plugin-name>` - Plugin identifier (matches plugin.json id)
- `<tool-name>` - Descriptive snake_case tool name

### Real-World Examples

```typescript
// Frontend Plugin
"mcp__frontend__figma_fetch"           // Fetch Figma designs
"mcp__frontend__figma_export_assets"   // Export Figma assets
"mcp__frontend__lighthouse_audit"      // Run Lighthouse audit

// Code Analysis Plugin
"mcp__code-analysis__claudemem_search"       // Search codebase
"mcp__code-analysis__claudemem_enrich"       // Enrich file context

// Bun Backend Plugin
"mcp__bun__apidog_sync"                // Sync with Apidog
"mcp__bun__apidog_validate"            // Validate API spec

// SEO Plugin
"mcp__seo__analyze_page"               // Analyze page SEO
"mcp__seo__check_schema"               // Validate schema markup
```

### Tool Name Guidelines

**DO**:
- Use snake_case for tool names
- Use action verbs (fetch, create, update, analyze)
- Be specific about what the tool does
- Keep names under 50 characters

**DON'T**:
- Use camelCase or PascalCase
- Use generic names like "do_thing"
- Include version numbers in names
- Use abbreviations unless widely known

### Verb Conventions

| Ver