A Beginner's Guide to AI Agent Skills and Custom Tools

A bare AI agent can read and write text. That's it. To do anything useful — search the web, query a database, send an email — it needs skills.

Here's what skills are, how they work, and how to give your agent more of them.

What a Skill Is

A skill is a bundle of one or more tools that your agent can call. Each tool has:

• A name (e.g., web_search)
• A description (what it does, in plain English the model can understand)
• An input schema (what arguments it takes)
• A handler (the actual code that runs when the tool is called)

When the agent decides it needs to search the web, it calls web_search({"query": "..."}) and gets results back. Simple as that.

Built-in Skills in Noomachy

Noomachy ships with a curated set of built-in skills:

• Web Search — Google-style search via Serper
• File Operations — read/write files in cloud storage
• Code Execution — run JavaScript or Python in a sandbox
• Database Query — read from your Firestore data
• Calculator — math, conversions, date arithmetic
• JSON Tools — parse, transform, format
• Wikipedia — search articles
• Weather — current conditions and forecasts
• Text Analysis — sentiment, entities, summarization
• Image Generation — text-to-image

Plus, when you install the desktop app, you get:

• Mail, Notes, Calendar, Reminders — Apple ecosystem integration
• Clipboard, Files, System — local Mac control

Browse the skill marketplace →

How to Install a Skill

In the Noomachy dashboard:

1. Open Skills in the sidebar
2. Browse the marketplace
3. Pick the skills you want
4. Click Install to add them to your selected agent

Installed skills become available tools the next time the agent receives a message.

How to Add a Custom MCP

Beyond built-in skills, you can connect any external MCP server (Model Context Protocol). Examples:

• A Notion MCP server that lets the agent read your databases
• A Linear MCP server for managing tickets
• A custom internal API exposed as MCP

In Noomachy:

1. Open Skills
2. Click Add Custom MCP
3. Enter the name, description, and HTTPS endpoint
4. Optionally add an API key
5. Click Add MCP Server

Your agent will discover the new tools from the MCP server's /tools/list endpoint and start using them immediately.

Building Your Own MCP Server

A minimal MCP server is just an HTTP service with two endpoints:

POST /tools/list  → returns tool definitions
POST /tools/call  → executes a named tool with arguments

You can write one in any language. Node, Python, Go, Rust — they all work because the protocol is just HTTP+JSON.

We've published the source for Noomachy's local MCP server as an example.

Tool Use Tips

• Less is more. Don't install every skill on every agent. The more tools an agent has, the more confused the model gets about which to use. Stick to 5-15 highly relevant tools per agent.
• Description matters. The model reads the tool description. Make it crisp and concrete.
• Schema enforcement. Use strict JSON schemas. Sloppy schemas = sloppy tool calls.

Try It

Browse the Skill Marketplace → or sign up for Noomachy to start installing skills.