AI Tools for Agents & MCP Servers

If you're building AI agents or MCP servers in Xano, tools are where everything begins. Before you wire up an agent or expose an MCP server, you need to understand what tools are, why they're built the way they are, and how to create one from scratch. Let's walk through the foundation.

What Is a Tool in Xano?

A tool is a scoped function stack — it takes defined inputs, performs one focused job, and returns an output. Think of tools as individual skills: you might build one to summarize a message, classify intent, fetch user info, or trigger any other discrete action your workflow needs.

Tools live independently from agents and MCP servers for three important reasons:

• Reusability — the same tool can be assigned to multiple agents or MCP servers
• Testability — you can run and debug a tool in isolation without spinning up a full agent
• Clean architecture — tools are atomic; agents handle orchestration, and MCP servers act as the gateway

A helpful mental model: tools are skills, agents are logic, and MCPs are access.

Navigating the Tools UI

To get started, head to the AI tab in Xano and select Tools. Here you'll see a list of all your existing tools, including their title, tags, where they're used (an MCP server, an agent, or both), and when they were created and published.

Creating Your First Tool

Click Add Tool in the top right corner (or the blue button on a blank canvas) to create a new tool. Give it a name — for example, summarize_message — and add a clear description. You can also apply tags to keep things organized, such as "summarization."

The most important field here is Tool Instructions. This tells AI agents and MCP servers how and when to use the tool, so be descriptive and intentional.

Once saved, you'll enter the function stack editor, which works just like the rest of the Xano platform. Add your inputs, build your logic, and define your response. For the summarize tool, you'd add a text input called input, then drop in a custom function that calls the OpenAI completions API with a system prompt instructing it to summarize text into one sentence.

Testing Before You Deploy

One of the biggest advantages of tools is that you can test them directly in the UI using the Run and Debug button — no agent or MCP server required. Pass in a sample input, run it, and verify the output looks right before publishing.

Once you're satisfied, hit Publish, and the tool becomes available to assign to any agent or MCP server.

Managing Your Tools

The three-dot More Options menu gives you access to several useful features: editing settings and instructions, reviewing version history for rollback flexibility, checking request history for logging, cloning a tool as a starting point for a new one, and converting a tool into a reusable function if your needs change.

Tools also respect any middleware you configure at the agent or MCP server level, so authentication and access controls flow through automatically.

With your first tool built and published, you're ready to start assembling agents that put these skills to work.