MCP Server: SaaS Gateway
Provides tools for interacting with Stripe's REST API, enabling listing customers and optionally other operations.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MCP Server: SaaS Gateway list stripe customers and postmaid profile"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
π MCP Server: SaaS Gateway
A small TypeScript template for building one Model Context Protocol server that fronts many REST APIs.
One gateway, N apps. Drop tokens in env vars, expose each API as a slug, ship a single MCP entry to Claude Desktop / Claude Code.
Why this exists
The default pattern in the MCP ecosystem is one server per backend. That works fine when you only need Slack OR GitHub. It gets messy fast when you actually want Claude to orchestrate across several of your own SaaS APIs in a single conversation β every backend becomes another entry in claude_desktop_config.json, with its own token to rotate.
This template flips the model: one gateway server, N apps. Each app is a slug + baseUrl + Bearer token. Tools are named {app_slug}_{action} so Claude understands which API does what.
Born from a real production setup that fronts 4 internal SaaS (PostMaid, Daiv, Botlers, AGSteel) and exposes ~30 tools through one MCP entry β total config footprint in Claude Desktop: 6 lines.
This is not a published npm package β it's a template. Fork it, edit src/index.ts and src/config.ts, ship your own gateway.
Related MCP server: InfraClaude
Quick start
# 1. Clone the template
git clone https://github.com/grobinson3108/mcp-server-saas-gateway.git my-gateway
cd my-gateway
# 2. Install + build
npm install
npm run build
# 3. Run with your tokens
EXAMPLE_API_TOKEN=your-token npm startWire into Claude Desktop:
{
"mcpServers": {
"my-gateway": {
"command": "node",
"args": ["/absolute/path/to/my-gateway/dist/index.js"],
"env": {
"EXAMPLE_API_TOKEN": "your-token"
}
}
}
}Restart Claude Desktop. Your tools appear under the π icon.
What's inside
File | Role |
| Tiny generic HTTP client. Bearer auth, JSON in/out, consistent error format. |
| Config schema + |
| The entry point β also the example. This is the file you edit to register your apps and tools. |
| Real-world example: 4 apps, ~30 tools. |
| Smallest possible setup: one third-party API, two tools. |
The pattern
// 1. Declare your apps
const config = loadConfigFromEnv('my-gateway', [
{ slug: 'postmaid', name: 'PostMaid', defaultBaseUrl: '...', type: 'internal' },
{ slug: 'stripe', name: 'Stripe', defaultBaseUrl: 'https://api.stripe.com/v1', type: 'external' },
]);
// 2. Build a client per app
const clients = Object.fromEntries(
Object.entries(config.apps).map(([slug, app]) => [slug, new ApiClient(app)])
);
// 3. Register tools, namespaced by slug
server.tool('postmaid_get_profile', 'Get PostMaid editorial profile', {}, async () => {
return wrap(await clients.postmaid.get('/profile'));
});
server.tool('stripe_list_customers', 'List recent Stripe customers',
{ limit: z.number().optional() },
async ({ limit }) => wrap(await clients.stripe.get('/customers', { limit: String(limit ?? 10) }))
);That's it. No registry magic, no codegen, no schema crawling. Each tool is a 3-line lambda that calls one endpoint.
When to use this template vs alternatives
Setup | Best for |
This template | You own / use several REST APIs and want Claude to orchestrate across them. |
One MCP server per API (punkpeye/awesome-mcp-servers) | You only need one backend and want to use someone else's published server. |
Anthropic TS SDK raw | You're building something significantly more complex than HTTP gateway. |
laravel-mcp-server (PHP/Laravel) | Your app is the backend and is built in Laravel. |
Trade-offs we made
No schema introspection / OpenAPI codegen. Adding it would buy you 10 minutes per tool while costing a lot of maintenance complexity. Tools are 3 lines of TypeScript β writing them by hand stays fast and readable.
stdio only. HTTP/SSE transport is on the roadmap if a real use case shows up.
No per-tool authorization. All apps share the gateway's spawn-time env. If you need per-call scopes, do it inside the tool callback (the Bearer token of the app is yours to scope however you want).
Not on npm. This is a starter template β fork-and-edit beats install-and-configure for this kind of code.
Examples
examples/multi-saas-gateway/β A 4-app gateway inspired by a real production setupexamples/stripe-only/β Minimal Stripe wrapper
License
MIT β see LICENSE.
About the author
Greg Robinson β AI Architect, RAG and agentic systems, Audelalia (π«π· Montpellier).
Companion package for the PHP side: laravel-mcp-server β same idea, but for Laravel apps where the SaaS itself is the backend.
β If this saved you an hour, drop a star.
Available Tools
2 toolsexample_get_resourceB
Fetch a resource from the Example API by ID.
Inputs: resource_id (required).
gateway_list_appsA
List all SaaS apps wired into this gateway, with their type and configuration status.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogβ¦
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: one fetches a resource by ID, the other lists all apps configured in the gateway. There is no overlap in functionality.
Both tools follow a consistent verb_noun pattern using snake_case (example_get_resource, gateway_list_apps), maintaining predictable naming.
With only 2 tools, the server is at the low end of the borderline range. While not extreme, it feels thin for a SaaS gateway that would typically require more operations.
The tool set is missing essential CRUD operations for resources, app management (create, update, delete), and configuration details. Agents would fail when trying to perform common gateway tasks.
Maintenance
Related MCP Connectors
One MCP endpoint for Claude, GPT & Gemini: 100+ tools + no-code connectors + agent workers.
MCP server unifying ERPs, CRMs, APIs and knowledge base for Claude, ChatGPT and Gemini.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
The Remote MCP server acts as a standardized bridge between LLM applications (like Claude, ChatGPT, and Cursor) and external services, enabling AI agents to access external tools and resources. Its primary capability is providing a centralized search tool to discover other MCP servers and their respective tools. Unlike local implementations, it runs remotely with OAuth authentication and permission controls for security.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA generic MCP server that dynamically converts OpenAPI-defined REST APIs into tools for LLMs like Claude. It supports multiple authentication methods and transport protocols, enabling seamless interaction with any OpenAPI-compliant API.15 npmMIT
- AlicenseBqualityDmaintenanceAn MCP server that enables Claude to manage infrastructure across Kubernetes, Docker, Prometheus, and Terraform through natural language. It provides over 42 specialized tools with a safety-first design, including risk-based command classification and audit logging.43MIT
- FlicenseNot gradedqualityDmaintenanceA unified data gateway MCP server that enables Claude to interact with multiple external REST and GraphQL APIs through OAuth 2.0 authentication.-
- AlicenseAqualityDmaintenanceOne MCP server for the SaaS back office. Stripe, HubSpot, and Google Sheets exposed as typed, read-only-by-default tools for Claude and any MCP client.11MIT