MCP Production Kit

The enterprise-ready, type-safe TypeScript starter kit for building Model Context Protocol (MCP) servers.
Built for Claude Desktop, Cursor, Antigravity, and autonomous AI agent architectures.

---

Free Core vs. Pro Pack Comparison

Capability	Free Open-Source Core	Commercial Pro Pack ($19)
MCP Spec Compliance	Full (v2024-11-05 JSON-RPC 2.0)	Full (v2024-11-05 JSON-RPC 2.0)
Local Stdio Transport	Included (Claude / Cursor)	Included
HTTP / SSE Streaming Transport	Included (npm run start:sse)	Included
Diagnostic Tool	get_system_metrics (OS/Memory/Load)	Included
Security Tool	secure_http_fetch (SSRF Guard)	Included
State Store Tool	agent_memory_store (Namespaced KV + TTL)	Included
Logging Pipeline	Zero-leak stderr streaming	Zero-leak stderr streaming
Glama Registry Listing	Listed on Glama Registry	Listed on Glama Registry
Edge Serverless Deployment	-	Cloudflare Workers template (global edge network)
Database Connector	-	PostgreSQL / Supabase with token & keyword mutation guards
Container Support	Standard Dockerfile	Production Docker Compose (Fly.io / Railway / AWS)
Commercial Distribution	Free under MIT License	Commercial License + Instant ZIP Download

---

Why MCP Production Kit?

Most open-source MCP examples are proof-of-concepts: they demonstrate a toy tool over local stdio, but fail when faced with real-world deployments.

MCP Production Kit solves the core engineering hurdles out of the box:

• Dual Transport Architecture: Run locally on stdio (for Claude Desktop / Cursor) or remotely over HTTP / Server-Sent Events (SSE) using a single codebase.
• Zero-Leak Logging: In stdio mode, stdout is strictly reserved for JSON-RPC protocol framing. Diagnostics are piped to stderr with automated secret redacting to prevent pipe corruption.
• Hardened Security & SSRF Defense: The built-in HTTP fetch tool blocks loopback addresses, private IPv4/IPv6 ranges, and cloud metadata endpoints (169.254.169.254).
• In-Memory Agent Persistence: Includes an isolated, namespaced key-value state store with configurable TTL for multi-step agent reasoning.
• Production Observability: Built-in OS telemetry, memory metrics, and /health monitoring endpoint.

---

Architecture Overview

+-------------------------------------------------------------+
|                AI Agent (Claude / Cursor)                   |
+-------------------------------------------------------------+
                              |
                 JSON-RPC 2.0 / MCP Protocol
                              |
+-------------------------------------------------------------+
|                      Transport Layer                        |
|   +-----------------------+     +-----------------------+   |
|   |    stdio (Desktop)    |     |      HTTP / SSE       |   |
|   +-----------------------+     +-----------------------+   |
+-------------------------------------------------------------+
                              |
+-------------------------------------------------------------+
|                    MCP Production Server                    |
|   +-----------------------+     +-----------------------+   |
|   |     Zod Validator     |     |  Safe Stderr Logger   |   |
|   +-----------------------+     +-----------------------+   |
|                                                             |
|   Built-in Core Tools:                                      |
|   - get_system_metrics                                      |
|   - secure_http_fetch (SSRF Guard)                          |
|   - agent_memory_store (Namespaced KV + TTL)                |
+-------------------------------------------------------------+

---

Quickstart

1. Installation

git clone https://github.com/A7mdMe/mcp-production-kit.git
cd mcp-production-kit
npm install
npm run build

2. Run Verification Tests

npm test

3. Connect to Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "production-kit": {
      "command": "node",
      "args": [
        "/path/to/mcp-production-kit/dist/index.js",
        "--transport=stdio"
      ]
    }
  }
}

4. Run as Remote HTTP/SSE Server

npm run start:sse
# Listening on http://localhost:3000
# Healthcheck: http://localhost:3000/health
# SSE Stream: http://localhost:3000/sse

5. Run via Docker

docker build -t mcp-production-kit .
docker run -p 3000:3000 -e TRANSPORT=sse mcp-production-kit

---

Built-in Tools Specification

Tool Name	Description	Security Controls
get_system_metrics	Real-time OS, CPU load, and heap memory breakdown.	Read-only introspection.
secure_http_fetch	Safe web scraping and API interaction.	Blocks private IPs, AWS metadata, caps payload at 512 KB, 10s timeout.
agent_memory_store	Multi-turn state management (set, get, list, delete).	Isolated namespaces and automatic TTL expiration.

---

Commercial Pro Pack ($19 One-Time)

For production cloud deployments and enterprise teams, the Pro Pack provides ready-to-deploy extensions:

• Cloudflare Workers Serverless Template: Low-latency edge deployment with Bearer Token auth and native streaming.
• PostgreSQL / Supabase Connector: Read-only schema introspection and parameterized SQL queries with comment-stripping and mutation keyword defenses.
• Production Docker Compose: Multi-stage container with non-root security, healthcheck polling, and reverse proxy setup.

Get Pro Pack ($19.00 USD) — Instant Checkout on Whop
(Delivery: Instant download via Whop Files app and customer billing portal upon completion).

---

License

MIT © 2026 Antigravity Autonomous Lab