Kaggle Plugin for Dify

Author: ki3dn
Type: tool
Github Repo: https://github.com/ki3nd/kaggle-dify-plugin
Github Issues: issues

Run and manage Kaggle kernels directly from your Dify workflows. Write Python code, choose your accelerator, poll for completion, and retrieve output files — all without leaving your AI pipeline.

Setup

Go to kaggle.com → Settings → API → Create New API Token.
Copy the token value (a long alphanumeric string).
In Dify, open the plugin settings and paste the token into the KAGGLE_API_TOKEN credential field.

Tools

List User Kernels

List kernels owned by any Kaggle user, or by the authenticated account when no username is supplied.

Parameter	Required	Description
	No	Kaggle username. Leave empty to list your own kernels.

Returns: a list of entries.

Create Kernel

Create a new Kaggle kernel with a minimal Python starter script. The kernel slug is derived automatically from the title.

Required	Default	Description
Yes	—	Kernel title (minimum 5 characters). Must be unique among your kernels.
No		Whether the kernel is private.
No		Whether the kernel has internet access during execution.

Returns: kernel metadata including , , and settings.

Run Kernel Code

Push Python code to an existing kernel and trigger a new run. The kernel is converted to a Python script type before pushing, so the provided code replaces the current script entirely.

Required	Default	Description
Yes	—	or just (auto-prefixed with your username).
Yes	—	Python source code to run.
No		, , , or .

Returns: push result including , , and any errors.

Get Kernel Status

Fetch the latest run status and full metadata for a kernel.

Parameter	Required	Description
	Yes	or just (auto-prefixed with your username).

Returns: .

Possible status values: , , , , .

Get Kernel Metadata

Fetch the configuration metadata for a kernel (language, kernel type, datasets, accelerator, etc.) without checking its run status.

Parameter	Required	Description
	Yes	or just (auto-prefixed with your username).

Returns: full kernel metadata object.

If the kernel does not exist or is private/inaccessible, a descriptive message is returned instead of raising an error.

Get Kernel Output

Download a specific output file from a completed kernel, or retrieve only the execution logs when no file path is given.

Parameter	Required	Description
	Yes	or just (auto-prefixed with your username).
	No	Path to the output file. Accepts absolute Kaggle paths () or relative paths (). Leave empty to return logs only.

Returns:

With : file content (text, JSON, image, or binary blob) followed by .
Without : — no file is downloaded.

If the kernel is still running or queued, an informative message is returned and the download is skipped. If the kernel failed, the failure message and status are returned.

Download Kernel Output

Download all output files from a completed kernel and return them as a single zip archive. Use this tool when you need all output files at once rather than a specific one.

Parameter	Required	Description
	Yes	or just (auto-prefixed with your username).

Returns: a zip blob named followed by .

If the kernel produced no output files, an informative message is returned along with the logs. Status checks (running, queued, failed) behave the same as Get Kernel Output.

Kernel ID shorthand

Every tool that accepts a supports a short form. If you omit the prefix, the plugin automatically prepends the authenticated user's username:

Typical workflow

Create Kernel once to get a persistent .
Run Kernel Code to push new code and trigger execution.
Get Kernel Status in a loop until .
Get Kernel Output to retrieve a specific result file or logs.
Download Kernel Output to retrieve all output files packaged as a zip.