mcp-helmet
A production middleware toolkit for MCP servers providing auth, health checks, graceful shutdown, and transport ergonomics, enabling rapid development of robust MCP servers.
README
mcp-helmet
Production middleware for MCP servers. Auth, sessions, health checks, graceful shutdown, transport ergonomics. Composable middleware borrowed in spirit from Express's
helmet.
mcp-helmet wraps the official @modelcontextprotocol/sdk with the things it doesn't ship: auto transport detection, content wrapping, health checks, graceful shutdown, session management, and auth middleware. One package. Composable. Drop what you don't need. Go from hello world to production in minutes, not days.
npm install mcp-helmet @modelcontextprotocol/sdk zod
Peer dependencies:
@modelcontextprotocol/sdk^1.29.0,zod^3.22.0 or ^3.25 (v4).zod-to-json-schemais an optional peer for Zod v3 users.
Quickstart
The fastest way is the scaffolder:
npx mcp-helmet init my-server --transport http --auth bearer
cd my-server
npm install
npm run dev
You get a working MCP server with healthCheck(), gracefulShutdown(), optional auth, a multistage Dockerfile, and a typechecked tsconfig. Drop flags to skip pieces (--no-docker, --no-health, etc.) or customise after the fact.
Or wire it manually:
import { createServer } from "mcp-helmet";
import { z } from "zod";
const server = createServer({ name: "hello", version: "1.0.0" });
server.tool("greet", { name: z.string() }, async ({ name }) => {
return `Hello, ${name}!`;
});
await server.start();
That's it. No transport wiring, no content array construction, no signal handlers. Run it:
# Local development (stdio, the default)
npx tsx src/index.ts
# Production (HTTP)
MCP_TRANSPORT=http PORT=3000 node dist/index.js
Same code, both modes.
Auth in 6 lines
Bearer token verification, with the verified principal available inside any tool handler:
import { createServer, bearerAuth, getAuthContext } from "mcp-helmet";
const server = createServer({ name: "secure", version: "1.0.0" });
server.use(bearerAuth({
verify: async (token) => {
const claims = await verifyJwt(token); // your call
return { user: claims.sub, scopes: claims.scope?.split(" ") ?? [] };
},
}));
server.tool("whoami", {}, async () => {
const auth = getAuthContext();
return { user: auth?.user, scopes: auth?.scopes };
});
await server.start();
The single-argument tool handler signature stays the same — getAuthContext() reads from AsyncLocalStorage, so it works from any depth in the async chain.
Why this exists
We audited 30 production MCP servers and 320 GitHub issues across the official SDKs. Three patterns kept showing up:
-
Every server rewrites the same 20-40 lines of setup. Transport selection, content wrapping, error formatting, signal handling. The SDK gives you the building blocks; this gives you the house.
-
Servers that work locally break in production. Docker containers exit after one response, Kubernetes pods lose sessions, no health check for load balancers to probe. 52% of remote MCP endpoints in a recent survey were dead.
-
Nobody can figure out auth. "How do I access the bearer token inside my tool?" is the most asked question across both SDK repos.
mcp-helmet solves these with composable middleware that extends the SDK without replacing it.
Status
v0.1.0-alpha — feature complete. Currently shipped:
- ✅
createServer()with auto content wrapping (string, object, Content[]) - ✅ Auto transport detection via
MCP_TRANSPORTenv var - ✅ Zod v3 + v4 compatibility shim
- ✅ Composable middleware system (
server.use(mw)) - ✅
healthCheck()middleware - ✅
gracefulShutdown()middleware - ✅
bearerAuth()andapiKeyAuth()middleware with AsyncLocalStorage-basedgetAuthContext() - ✅
rateLimiter()middleware (sliding window, IP- or key-based, standard 429 headers) - ✅
npx mcp-helmet initCLI scaffolder + Docker template
v0.1.0 stable will follow once the alpha cycle has 30+ days of real-world usage. v0.2 is on the ROADMAP: Redis-backed session store, structured logging middleware, Python port via FastMCP plugin.
How it relates to the official SDK
mcp-helmet is not a fork, an alternative, or a replacement. It's a convenience layer.
| Concern | Official SDK | mcp-helmet |
|---|---|---|
| Protocol implementation | Yes | No, delegates to SDK |
| Transport classes | Yes | No, wraps SDK transports |
| Tool/resource/prompt registration | Yes | Yes, thinner API |
| OAuth server flows | Yes (in v2 dev) | No, out of scope |
| Bearer/API-key middleware | Express-coupled in v2 | Transport-agnostic, composable |
| Health checks | No | Yes, planned |
| Session externalization | No | Stopgap until upstream SEP |
| Docker/deployment templates | No | Yes, planned |
The SDK is a peer dependency. You bring your own version. If the SDK adds features that overlap, the toolkit middleware becomes a thin pass-through.
Requirements
- Node.js 20+
@modelcontextprotocol/sdk^1.29.0- TypeScript 5.4+ (recommended, not required)
- Zod 3.22+ or 3.25+ (v4 via
zod/v4subpath)
Contributing
PRs and issues welcome. See CONTRIBUTING.md for setup, test, and PR conventions.
Security
See SECURITY.md for the responsible disclosure path.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.