# MCP server tutorial: connect an app to an AI agent

> A short MCP server tutorial for connecting a supported app to an AI agent through a hosted connection, then checking that the agent can discover its tools.

Web version: https://claw-link.dev/learn/mcp-server-tutorial

This tutorial connects a supported app to an MCP-capable agent without building
an OAuth application or running a server yourself. It uses ClawLink's hosted
connection flow. For an internal API or custom tool, use the same sequence as a
model for planning a self-hosted server, but own the server and authorization
steps yourself.

## Setup

1. Install and sign in to the ClawLink CLI:

```bash
npx -y @useclawlink/cli login
```

2. Install the connection for your agent. For example, the CLI can install its
MCP configuration for a supported coding agent:

```bash
npx -y @useclawlink/cli install <agent>
```

3. Start a browser connection for an app such as Gmail:

```bash
npx -y @useclawlink/cli connect gmail
```

4. Complete the provider consent screen. Review the account and permissions
before approving.
5. Return to the agent and ask it to list the app's available actions or do a
read-only task such as finding a recent email.

The browser approval connects the app. The agent configuration tells the agent
where to reach the MCP tools. Those are separate steps, and skipping either one
produces a different failure.

## Verify the connection

Start with a read. For Gmail, ask for a recent subject line. For Google Drive,
ask the agent to find a file by name. A successful read proves that the agent
can discover the server, select a tool, and use the connected account.

Do not make a send, delete, or publish action your first test. A read gives you
the same connection signal without changing an external system.

## If the agent cannot use the app

Check these in order:

1. The app connection says connected in ClawLink.
2. The agent configuration points to the MCP server installed by the CLI.
3. The agent has started a fresh session or reloaded its tool catalog.
4. The requested tool exists and the account can do the same action in the app
   UI.
5. The provider consent covered the permission needed for the action.

The [MCP tool call failed](https://claw-link.dev/learn/mcp-tool-call-failed) guide separates tool
discovery, arguments, permissions, provider failures, and re-authentication.

## When to use a different approach

This hosted tutorial is not for an API that only exists inside your company
network or for a tool whose behavior is part of your own product. In those
cases, build or operate a self-hosted MCP server and use the authorization and
architecture guides below.

## Keep reading