Skip to content

Connect your own tools with MCP ​

By the end of this tutorial your agent won't just answer — it will act: look up an order, check a booking, charge a card — using tools you own, running against your own systems.

This is a hands-on lesson and the natural next step after your agent can talk and answer from documents. It assumes you (or a developer on your team) have an MCP server, or that you'll connect one of the ready-made providers from the catalog to follow along.

What you'll build ​

An agent with an Integration connected, so the actions your MCP server exposes become tools the agent can call during a conversation.

Before you start ​

  • An agent you can edit.
  • An MCP server — your own, or a provider from the built-in catalog to practise with. An MCP server is where your business logic and data live; Perfox calls its tools but never holds your domain logic itself. New to the idea? Read Agents and MCP first — it's a five-minute concept.

1. Connect the server ​

Go to Connect → Integrations. You have two ways to add a server:

  • On the Browse tab, click Connect on a ready-made provider and fill in what it asks for (or sign in, for an OAuth provider), or
  • On the Browse tab, click Add custom server and enter your server's name, URL, Transport and Auth Type (with its key or token, if it needs one), then click Add Integration.

Either way, Perfox connects to the server and reads the actions (tools) it offers.

You should now see the server under the Connected tab, marked Healthy, with a tool count. Click the card to expand it and see every tool and its description.

2. Attach it to your agent ​

Open your agent on the canvas. On the AI Agent node, click the + under Integration to add an Integration node, open it, and pick the server you just connected. Under Enabled actions (optional), choose the actions this agent may use.

Why this matters. The Integration port is the bridge between the agent and your systems. Exposing only the actions this agent needs keeps it focused — and cheaper, because every exposed action's description is sent to the AI on every turn. If your server adds a tool later, click Rediscover on its card in Integrations before you can pick it.

3. Tell the agent when to use the tools ​

Click the Personality port and describe when to reach for a tool, in plain language:

When a customer asks about an order, use the order-lookup tool with their order number before answering. Never guess an order status — always read it from the tool result.

Then click Activate.

4. See it act ​

Open your agent's live preview and ask something that needs a real lookup — "What's the status of order #4821?"

You should now see the agent call your tool, get the real result, and answer with actual data — not a guess. Every tool call is recorded, so you can trace exactly what the agent asked for and what came back.

What just happened. The agent decided, mid-conversation, that it needed your order-lookup tool, called it with your credential attached, and answered from the result. Your domain logic stayed entirely in your own server — Perfox routed the call and recorded it in the conversation.

You did it ​

Your agent can now take real actions against your own systems, safely and traceably. Where to go next: