> ## Documentation Index
> Fetch the complete documentation index at: https://docs.flow-board.co/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP

> Connect AI tools to Flowboard with MCP to read live flows, edit visual-editor drafts, and inspect analytics.

> Connect AI tools to Flowboard with the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction).

Flowboard MCP gives compatible AI clients access to your Flowboard app so they can inspect live flows, work against the visual editor draft, and read analytics.

<Warning>
  Flowboard MCP requires an API key. In the Flowboard dashboard, open your app, then go to **App settings** → **Integrations** to create an MCP key.
</Warning>

## Connect

* MCP URL: `https://mcp.flow-board.co`
* Auth: `Authorization: Bearer <your_mcp_key>`
* MCP keys are generated with `flows:read`, `flows:write`, and `stats:read`

Most remote MCP clients accept a config like this:

```json theme={null}
{
  "mcpServers": {
    "flowboard": {
      "url": "https://mcp.flow-board.co",
      "headers": {
        "Authorization": "Bearer <your_mcp_key>"
      }
    }
  }
}
```

## Setup by tool

<Info>
  Flowboard works best in MCP clients that let you send a custom `Authorization` header directly.
</Info>

<Accordion title="Cursor">
  Create `.cursor/mcp.json` in your project, or `~/.cursor/mcp.json` for a global setup.

  ```json theme={null}
  {
    "mcpServers": {
      "flowboard": {
        "url": "https://mcp.flow-board.co",
        "headers": {
          "Authorization": "Bearer <your_mcp_key>"
        }
      }
    }
  }
  ```
</Accordion>

<Accordion title="VS Code">
  Create `.vscode/mcp.json` in your workspace.

  ```json theme={null}
  {
    "servers": {
      "flowboard": {
        "type": "http",
        "url": "https://mcp.flow-board.co",
        "headers": {
          "Authorization": "Bearer <your_mcp_key>"
        }
      }
    }
  }
  ```
</Accordion>

<Accordion title="Windsurf">
  Open **Windsurf Settings** → **Tools** → **Windsurf Settings** → **Add Server**. For raw config, edit `~/.codeium/mcp_config.json`.

  ```json theme={null}
  {
    "mcpServers": {
      "flowboard": {
        "serverUrl": "https://mcp.flow-board.co",
        "headers": {
          "Authorization": "Bearer <your_mcp_key>"
        }
      }
    }
  }
  ```
</Accordion>

<Accordion title="Antigravity">
  Open the agent panel, click `...` → **MCP Servers** → **Manage MCP Servers** → **View raw config**, then update `mcp_config.json`.

  ```json theme={null}
  {
    "mcpServers": {
      "flowboard": {
        "serverUrl": "https://mcp.flow-board.co",
        "headers": {
          "Authorization": "Bearer <your_mcp_key>"
        }
      }
    }
  }
  ```

  If your Antigravity build expects `url` instead of `serverUrl`, use the same endpoint under `url`.
</Accordion>

## What Flowboard exposes

Flowboard MCP exposes product capabilities, not raw database access. In practice, it is backed by these stores:

| Backing store or source                   | Exposed through MCP                                                                                 |
| ----------------------------------------- | --------------------------------------------------------------------------------------------------- |
| `onboardings`                             | Published flow metadata, live runtime JSON, localized screens, publish state                        |
| `onboardingDrafts`                        | Visual-editor draft project, draft versions, and draft updates                                      |
| `audiences`                               | Live flow resolution for an app/install context                                                     |
| `screenCatalogItems`                      | Built-in Flowboard templates plus team and app screen-library items used for generation and editing |
| `aiEditOperations` + `onboardings.aiLock` | Async edit status and draft locking for long-running editor edits                                   |
| BigQuery analytics dataset (`raw_events`) | Read-only app, flow, viewer, and experiment metrics                                                 |

For screen-library content, the AI layer pulls from:

* Flowboard templates: `screenCatalogItems` with `sourceType: flowboard`
* Team and app library items: `screenCatalogItems` with `sourceType: team`, filtered by `teamId` and optionally `sourceAppId`

## Editor-aware workflows

Flowboard MCP is aware of the Flowboard editor model:

* It can read the current draft and the published flow separately
* AI edits are draft-first and do not publish automatically
* Screen edits support a preview-first flow before apply
* Long-running edits are tracked and can temporarily lock the draft to avoid conflicting writes

Typical editor actions include:

* Generate a new flow
* Edit an existing draft flow
* Preview or apply a single-screen edit
* Generate a screen from an image
* Insert a generated screen into an existing flow
* Inspect flow analytics while iterating on the draft

## Example prompts

* "Fetch the draft version of flow `flow_123` and summarize each screen."
* "Preview a rewrite of the pricing screen to make it feel more premium, but do not apply it yet."
* "Insert a new testimonial screen before the paywall."
* "Show the last 30 days of funnel metrics for flow `flow_123`."