Skip to content

Getting Started

This page covers how to build an MCP server and connect it to Mentium. Once connected, AI Agents in Mentium will be able to use your server's tools to query data and take actions in your systems.

Suggested Starting Architecture

If you're building an MCP server for a freight organization and you want it to not turn into a mess:

1. Start with Read-Only Tools

Only later add tools that create or modify data. This lets you validate the AI Agent's behavior before giving it the ability to take actions with side effects.

2. Put All Side Effects Behind Tools

Every action that changes state should be an explicit tool:

  • Email send
  • CRM write
  • Ticket creation

This makes it easy to audit, scope, and gate actions.

3. Use API Keys at First

API keys work well for internal service-to-service MCP servers. Graduate to OAuth if you need per-user delegated access later (MCP's HTTP authorization spec is OAuth-based).

4. Require Human Approval for High-Risk Actions

Until your logging proves it's safe, keep a human in the loop for:

  • Sending customer-facing messages
  • Booking loads
  • Financial operations

5. Ship Observability from Day 1

You need visibility into what your agent is doing:

  • Tool call logs (what was called, with what inputs)
  • Failure tracking
  • Rate limiting metrics
  • Key usage statistics

Adding Your MCP Server to Mentium

To connect your MCP server to Mentium:

  1. Navigate to Settings -- Go to Settings > MCP Servers in the Mentium dashboard
  2. Add a new MCP server -- Click "Add MCP Server" and provide:
    • Server Name: A display name for your server
    • Server URL: The HTTPS endpoint of your MCP server (Streamable HTTP)
    • API Key: The API key for authenticating with your server
  3. Discover tools -- Mentium connects to your server, runs the MCP initialization handshake, and discovers available tools
  4. Enable/disable tools -- Choose which discovered tools should be available to AI Agents
  5. Use in agents -- The enabled tools automatically become available to AI Agents, giving them the ability to query your data and take actions in your systems

What's Next

  1. Create your MCP server -- Step-by-step guide with Python and TypeScript examples
  2. Deploy -- Host your server on HTTPS (required for production)
  3. Connect to Mentium -- Add it in Settings > MCP Servers
  4. Monitor -- Watch tool call logs and adjust scopes/permissions as needed

TIP

For more design guidance, see:

Mentium TMS API Documentation