@nebutra/mcp

Status: WIP

Status: WIP — Not yet integrated into any production app. Do not import until this notice is removed.

Model Context Protocol (MCP) for AI agent tool calling.

Installation

pnpm add @nebutra/mcp

Features

• MCP Client — Execute tools across registered servers
• Server Registry — Manage internal and external MCP servers
• Middleware — Rate limiting, audit logging, access control
• Plan-based Access — Tool access based on subscription tier

Architecture

Agent → MCPClient → ServerRegistry → MCP Servers
                  ↓
            Middleware (rate-limit, audit, access-control)

Usage

Initialize

import { mcpClient, registerInternalServers } from "@nebutra/mcp";

// Register built-in servers
registerInternalServers();

Execute Tools

const result = await mcpClient.executeTool(
  "generate_text",
  { prompt: "Write a product description" },
  {
    requestId: "req-123",
    tenantId: "org_456",
    plan: "PRO",
  },
);

Available Tools

| Tool | Plan | Description | | ----------------- | ---------- | ------------------- | | generate_text | FREE | AI text generation | | translate | FREE | AI translation | | embed_text | FREE | Generate embeddings | | recommend_items | PRO | Get recommendations | | sync_products | PRO | E-commerce sync | | mint_nft | ENTERPRISE | Web3 minting |

Plan-based Access Control

const plans = {
  FREE: ["generate_text", "translate", "embed_text", "content_*"],
  PRO: ["*_recommend_*", "*_ecommerce_*", ...FREE],
  ENTERPRISE: ["*_web3_*", ...PRO],
};

Register Custom Server

import { serverRegistry } from "@nebutra/mcp";

serverRegistry.register({
  name: "custom-tools",
  url: "http://localhost:3001/mcp",
  tools: [
    {
      name: "custom_tool",
      description: "My custom tool",
      parameters: { type: "object", properties: {} },
    },
  ],
  allowedPlans: ["PRO", "ENTERPRISE"],
});

Middleware

Rate Limiting

// Configured per tool
const limits = {
  generate_text: { requests: 100, window: "1m" },
  embed_text: { requests: 1000, window: "1h" },
};

Audit Logging

All tool executions are logged:

{
  "timestamp": "2024-01-15T10:30:00Z",
  "tool": "generate_text",
  "tenantId": "org_456",
  "userId": "user_123",
  "duration": 245,
  "success": true
}

Related

• AI service
• API Gateway