MCP Tools Deployment Template

A template for quickly creating and deploying your own MCP servers. It allows you to:

1. Quickly create completely custom MCP server just by chatting with AI IDEs.
2. Add pre-built open source MCP servers to your project from MCP marketplaces.
3. Create multiple MCP instances in separate directories with shared tools.

This template deploys all of them at once on railway in just a few clicks!

🚀 Step-by-Step Guide

Prerequisites

• Python 3.12+
• Git

Step 0: Copy Template

Copy this GitHub template by clicking the “Use this template” button on the top right of the page.

Step 1: Create a Virtual Environment

In Cursor, Windsur, or VS Code:

1. Open the Command Palette:
   • On Mac: Cmd + Shift + P
   • On Windows: Ctrl + Shift + P
2. Type and select: Python: Select Interpreter
3. Click: + Create Virtual Environment
4. Choose: Venv as the environment type

In Terminal:

python -m venv venv
source venv/bin/activate

Step 2: Create Your Tools

2.1 Create Custom Tools Just by Chatting With AI

Prompt your AI IDE to create the tools for you in chat. Make sure to include ./.cursor/rules/workflow.mdc in the context. (Included by default only in Cursor).

For example:

Please create a tool that fetches the transcripts from a YouTube video. @workflow.mdc

Answer the clarifying questions and keep iterating until the tools are created and working as expected.

Make sure to add any requested env variables to the ./.env file.

2.2 Add Pre Build Open Source Stdio MCP Servers

Add other pre-built Open Source Stdio MCP servers to the mcp.json file, similarly to Cursor and other clients.

{
  "mcpServers": {
    "notionapi": {
      "command": "npx",
      "args": [
        "-y",
        "@notionhq/notion-mcp-server"
      ],
      "env": {
        "OPENAPI_MCP_HEADERS": "{\"Authorization\":\"Bearer ntn_****\",\"Notion-Version\":\"2022-06-28\"}"
      }
    }
    # ... add more servers here
}

Only Stdio servers are supported by this method. This repo will automatically convert them to SSE with supergateway.

You can use both npx and uv to run these servers.

2.3 Test the MCP Servers (Optional)

Run the following command to test the MCP servers and ensure all of them are running.

python server/start_mcp.py

For testing with subdirectory endpoints:

python server/start_mcp.py --split-subdirs

You can’t connect local servers to Agencii, but you can test them by adding the SSE URL to Cursor’s “MCP Servers” tab.

This step is not necessary. As long as there are no issues in your tools.

Step 3: Multi-Endpoint MCP Server

This template supports serving multiple tool directories as separate endpoints from a single Python server.

Directory Structure