mcp-server-skills

Pattern for building MCP servers in Next.js with mcp-handler, shared Zod schemas, and reusable server actions.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "mcp-server-skills" with this command: npx skills add gocallum/nextjs16-agent-skills/gocallum-nextjs16-agent-skills-mcp-server-skills

Links

Folder Structure (Next.js App Router)

app/
  api/[transport]/route.ts   # One handler for all transports (e.g., /api/mcp)
  actions/mcp-actions.ts     # Server actions reusing the same logic/schemas
lib/
  dice.ts | tools.ts         # Zod schemas, tool definitions, pure logic
components/                  # UI that calls server actions for web testing

Goal: Keep route.ts minimal. Put logic + Zod schemas in lib/* so both the MCP handler and server actions share a single source of truth.

Shared Zod Schema + Tool Definition

// lib/dice.ts
import { z } from "zod";

export const diceSchema = z.number().int().min(2);

export function rollDice(sides: number) {
  const validated = diceSchema.parse(sides);
  const value = 1 + Math.floor(Math.random() * validated);
  return { type: "text" as const, text: `🎲 You rolled a ${value}!` };
}

export const rollDiceTool = {
  name: "roll_dice",
  description: "Rolls an N-sided die",
  schema: { sides: diceSchema },
} as const;

Reusable Server Actions (Web UI + Tests)

// app/actions/mcp-actions.ts
"use server";
import { rollDice as rollDiceCore, rollDiceTool } from "@/lib/dice";

export async function rollDice(sides: number) {
  try {
    const result = rollDiceCore(sides);
    return { success: true, result: { content: [result] } };
  } catch {
    return {
      success: false,
      error: { code: -32602, message: "Invalid parameters: sides must be >= 2" },
    };
  }
}

export async function listTools() {
  return {
    success: true,
    result: {
      tools: [
        {
          name: rollDiceTool.name,
          description: rollDiceTool.description,
          inputSchema: {
            type: "object",
            properties: { sides: { type: "number", minimum: 2 } },
            required: ["sides"],
          },
        },
      ],
    },
  };
}

Server actions call the same logic as the MCP handler and power the web UI, keeping responses aligned.

Lightweight MCP Route

// app/api/[transport]/route.ts
import { createMcpHandler } from "mcp-handler";
import { rollDice, rollDiceTool } from "@/lib/dice";

const handler = createMcpHandler(
  (server) => {
    server.tool(
      rollDiceTool.name,
      rollDiceTool.description,
      rollDiceTool.schema,
      async ({ sides }) => ({ content: [rollDice(sides)] }),
    );
  },
  {}, // server options
  {
    basePath: "/api",     // must match folder path
    maxDuration: 60,
    verboseLogs: true,
  },
);

export { handler as GET, handler as POST };

Pattern highlights

  • Route only wires createMcpHandler; no business logic inline.
  • server.tool consumes the shared tool schema/description and calls shared logic.
  • basePath should align with the folder (e.g., /api/[transport]).
  • Works for SSE/HTTP transports; stdio can be added separately if needed.

Claude Desktop Config (mcp-remote)

{
  "mcpServers": {
    "rolldice": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://localhost:3000/api/mcp"]
    }
  }
}

Best Practices

  1. Single source of truth — schemas + logic in lib/*; both MCP tools and server actions import them.
  2. Validation first — use Zod for inputs and reuse the same schema for UI + MCP.
  3. Keep route.ts light — only handler wiring, logging, and transport config.
  4. Shared responses — standardize { success, result | error } shapes for tools and UI.
  5. Vercel-friendly — avoid stateful globals; configure maxDuration and runtime if needed.
  6. Multiple transports — expose /api/[transport] for HTTP/SSE; add stdio entrypoint when required.
  7. Local testing — hit server actions from the web UI to ensure MCP responses stay in sync.

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Automation

prisma-orm-v7-skills

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

nextjs16-skills

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

authjs-skills

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

ai-sdk-6-skills

No summary provided by upstream source.

Repository SourceNeeds Review