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

# Model Context Protocol

> Install the Wood Wide MCP bundle to manage datasets, train models, and run inference from Claude Desktop.

Install the Wood Wide AI MCP bundle to manage datasets, train models, run inference, and download results from Claude Desktop.

The bundle connects Claude Desktop to your Wood Wide account using your API key. It can work with datasets and models already in your account, and it can upload local CSV or Parquet files when Claude Desktop has access to those files.

<CardGroup cols={2}>
  <Card title="Download MCP bundle" icon="download" href="https://console.woodwide.ai/settings?tab=mcp-connector">
    Get the `.mcpb` installer from the console.
  </Card>

  <Card title="API keys" icon="key" href="https://console.woodwide.ai/settings?tab=api-keys">
    Create or copy the API key used by the connector.
  </Card>
</CardGroup>

## Install the MCP bundle

<Steps>
  <Step title="Create an API key">
    The MCP server authenticates with your Wood Wide AI API key. [Find or create your API key](https://console.woodwide.ai/settings?tab=api-keys) before installing the bundle.
  </Step>

  <Step title="Download the .mcpb bundle">
    Open [Settings -> MCP connector](https://console.woodwide.ai/settings?tab=mcp-connector) and download the Wood Wide `.mcpb` file.
  </Step>

  <Step title="Open in Claude Desktop">
    Double-click the downloaded `.mcpb` file. Claude Desktop opens and starts installation.
  </Step>

  <Step title="Paste your API key">
    When prompted, paste your Wood Wide AI API key so the connector can reach your account.
  </Step>

  <Step title="Add the Filesystem MCP (recommended)">
    For end-to-end workflows that use local files, install the Filesystem MCP server from Claude Desktop's connectors list. This lets Claude read datasets from disk and save inference results locally.
  </Step>
</Steps>

## What You Can Do

Once installed, Claude Desktop can use Wood Wide tools to:

* List, inspect, and upload datasets
* List and inspect models
* Train prediction, clustering, anomaly, embedding, search, and factor models
* Run synchronous inference on a local file
* Start asynchronous batch inference
* Check async job status
* Download batch inference results to a local file

## Example Prompts

```text theme={null}
List my Wood Wide datasets and show the most recent ready dataset.
```

```text theme={null}
Upload ~/Downloads/customers.csv as a dataset, train a clustering model, then run inference on the same file.
```

```text theme={null}
Run batch inference with model [MODEL_ID] on dataset [DATASET_ID], wait for the job to finish, and save the results locally.
```

<Note>
  Claude Desktop only sees local files that your configured connectors can access. If a prompt mentions a file path, make sure the Filesystem MCP server is installed and allowed to read that location.
</Note>

***

## Related docs

* [API quickstart](/api-reference/getting-started) -- same workflow with curl, JavaScript, or Python
* [CLI](/cli) -- script datasets, training, and inference with `wwai`
* [SKILL.md](/skill-md) -- Cursor and Codex skill setup