# Build with your AI tool

> Connect Claude, ChatGPT, Cursor or any tool that supports MCP, then describe the automation you want. Your AI builds the workflow in Lodol, and Lodol runs it.

Web page: https://docs.lodol.com/docs/build-with-ai

Your AI tool is good at working out what an automation should do, but it can't keep one running. Connected to Lodol, it builds a workflow in your workspace instead of handing you a script, and Lodol runs that workflow on every trigger from then on, whether or not the AI tool is open.

Lodol connects through **MCP**, the open standard AI tools use to plug into other software. You sign in with your Lodol account when you connect, so there is no API key to copy.

```txt title="Lodol's MCP server"
https://mcp.lodol.com/mcp
```

## Connect your tool

<Tabs items={['Claude', 'ChatGPT', 'Cursor', 'Claude Code', 'VS Code', 'Other tools']} groupId="ai-tool" persist>
  <Tab value="Claude">
    [Add Lodol to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=Lodol&connectorUrl=https%3A%2F%2Fmcp.lodol.com%2Fmcp). Claude opens its connector settings with Lodol filled in. Click **Add**, then **Connect**, and sign in to Lodol.
  </Tab>
  <Tab value="ChatGPT">
    Turn on **Developer mode** in ChatGPT's settings. Then, under **Plugins**, create a plugin with the server URL above, choose **OAuth**, and sign in to Lodol.
  </Tab>
  <Tab value="Cursor">
    [Add Lodol to Cursor](https://cursor.com/install-mcp?name=lodol&config=eyJ1cmwiOiJodHRwczovL21jcC5sb2RvbC5jb20vbWNwIn0%3D). Cursor opens with Lodol ready to install. Click **Install**, then **Authenticate**, and sign in to Lodol.
  </Tab>
  <Tab value="Claude Code">
    Add the server from your terminal:

    ```bash
    claude mcp add --transport http lodol https://mcp.lodol.com/mcp
    ```

    Then run `/mcp` in Claude Code and choose **lodol** to sign in.
  </Tab>
  <Tab value="VS Code">
    Add Lodol to `.vscode/mcp.json` in your project, or to your user settings:

    ```json title=".vscode/mcp.json"
    {
      "servers": {
        "lodol": {
          "type": "http",
          "url": "https://mcp.lodol.com/mcp"
        }
      }
    }
    ```

    Start the server from the file and sign in to Lodol when VS Code asks.
  </Tab>
  <Tab value="Other tools">
    Add the server URL as a **remote MCP server** in Codex, Gemini CLI, Devin, Lovable, Replit, Zed, Cline or any other tool that supports MCP. Choose OAuth if the tool asks how to sign in; it then asks you to sign in to Lodol.
  </Tab>
</Tabs>

When you connect, you choose the workspace your AI works in, and you can untick any permission you don't want it to have.

## Ask for a workflow

Describe the automation the way you'd brief a colleague: what should start it, what should happen, and where things go.

```txt title="Try this in your AI tool"
Use Lodol: when a Stripe payment comes in, add it to our revenue sheet
and post it in #sales.
```

A few details make the first version right:

- **Say what starts it.** "When a payment comes in" asks for a trigger; "every Monday at 9:00" asks for a schedule. See [Triggers and schedules](https://docs.lodol.com/docs/concepts/triggers.md).
- **Name the places.** The sheet, the channel, the people to ask. Your AI leaves out anything it doesn't know rather than guessing, and tells you what to fill in.
- **Change it the same way.** "Only post payments over $1,000" updates the workflow you already have.

## What happens next

<Steps>
<Step>

### Your AI builds it and gives you a link

The workflow is saved in your Lodol workspace. Lodol checks it as it's saved, so it only uses apps, actions and fields that exist, and your AI tells you anything it left for you to fill in, such as a sheet it wasn't given.

</Step>
<Step>

### Connect the apps it uses

Open the link. Any app the workflow uses that isn't connected yet shows a **Connect** button at its step. This is a one-time setup per app; see [Connections and secrets](https://docs.lodol.com/docs/concepts/connections.md).

</Step>
<Step>

### Turn it on

From then on the workflow runs on its trigger or schedule. Each run is recorded step by step, and you get an email if one fails.

</Step>
</Steps>

## What your AI can do, and what it can't

Connected to Lodol, your AI can:

- find and explain the workflows in your workspace
- build new workflows and change existing ones
- run a workflow and read what happened in a run
- work out why a run failed and fix the workflow

It can't connect apps, see the values of your secrets or API keys, or manage members and billing. It can only do what your own [role](https://docs.lodol.com/docs/concepts/workflows.md#who-can-do-what) allows, and it can delete a workflow or a run only if you ticked **Delete workflows** or **Delete runs** when you connected it.

<Callout title="Every change can be undone">
  Each change your AI saves becomes a new [version](https://docs.lodol.com/docs/concepts/workflows.md#every-change-is-a-version) you can roll back to. Saving a change pauses the workflow until it's turned back on, which your AI can do for you.
</Callout>

## What it costs

Your AI tool only uses its own tokens while it builds or changes a workflow. Once the workflow exists, its runs come out of your Lodol plan, which counts each [run](https://docs.lodol.com/docs/concepts/runs.md) once however many steps it has, and no model is called unless a step asks for one.

You can disconnect your AI at any time from the **Connect AI** page in Lodol.