From Dashboard to AI Automation

Building an MCP Server for PurelyMail

Guilherme Fontes

September 23, 2025

From Dashboard Frustration to AI Automation

Building an MCP Server for PurelyMail

Problem: Manual email management friction

Solution: AI-powered automation via MCP

Result: Closer to Human-thinking-like account operations

The Problem: Death by Dashboard Clicks

  • Adding users → 7 clicks
  • Managing app passwords → 5 clicks
  • Domain configuration → 10+ clicks
  • Repetitive tasks → Mounting frustration

What is MCP?

Model Context Protocol = API to an API

AI Assistant ←→ MCP Server ←→ PurelyMail API
  • Standardized interface for AI tool integration
  • Type-safe operations with automatic validation
  • Natural language commands for complex tasks

Solution Evolution

Manual Clicks → Direct Scripts → MCP Abstraction

The Discovery: - Found PurelyMail’s OpenAPI specification - Automated client generation from swagger - Built systematic, reproducible methodology

Discovery Process

Step 1: Found the OpenAPI specification

https://news.purelymail.com/api/swagger-spec.js

Step 2: Automated everything - Type generation from swagger - Client code generation - Tool definitions creation

Build Process

graph LR
    A[Swagger Spec] --> B[Generate Types]
    B --> C[Generate Client]
    C --> D[Create MCP Tools]
    D --> E[Test & Deploy]

Commands:

npm run update:api    # Fetch + generate everything
npm run inspector     # Test tools locally

Code Examples

Tool Definition:

{
  name: "manage_user",
  description: "Create, delete, modify users",
  inputSchema: UserManagementSchema
}

Usage:

await client.manage_user({
  action: "Create User",
  userName: "newuser",
  domainName: "example.com"
});

Live Demo Time

What we’ll see: - MCP tools in action - Real PurelyMail account operations - Behind-the-scenes code generation - Manual vs automated comparison

Switching to live demo…

Results & Impact

Time Saved: - Manual process: ~2 minutes per user - MCP automation: ~10 seconds per user - 12x speed improvement

Beyond Speed: - Natural language interface - Transferable methodology

Key Takeaways

  1. Find OpenAPI specs - they’re a door for automation
  2. Generate, don’t write - let tools create boilerplate
  3. MCP abstraction > direct API scripting
  4. Automatic methodology beats one-off solutions

Food for thought

Your turn: What manual processes are begging for this treatment?

who knows… maybe you will have a “shower-idea”…

Thank You

Questions?

GitHub Repository

All Resources