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

1. Go to kaggle.com → Settings → API → Create New API Token.
2. Copy the token value (a long alphanumeric string).
3. 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
username	No	Kaggle username. Leave empty to list your own kernels.

Returns: a list of { kernel_id, kernel_name } entries.

---

Create Kernel

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

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

Returns: kernel metadata including id, url, 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.

Parameter	Required	Default	Description
kernel_id	Yes	—	owner/kernel-slug or just kernel-slug (auto-prefixed with your username).
code	Yes	—	Python source code to run.
accelerator	No	None	None, NvidiaTeslaP100, NvidiaTeslaT4, or TpuV5E8.

Returns: push result including url, version_number, and any errors.

---

Get Kernel Status

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

Parameter	Required	Description
kernel_id	Yes	owner/kernel-slug or just kernel-slug (auto-prefixed with your username).

Returns: { status, kernel_id, status_details, metadata }.

Possible status values: queued, running, complete, error, cancelled.

---

Get Kernel Metadata

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

Parameter	Required	Description
kernel_id	Yes	owner/kernel-slug or just kernel-slug (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
kernel_id	Yes	owner/kernel-slug or just kernel-slug (auto-prefixed with your username).
file_path	No	Path to the output file. Accepts absolute Kaggle paths (/kaggle/working/output.csv) or relative paths (images/result.png). Leave empty to return logs only.

Returns:

• With file_path: file content (text, JSON, image, or binary blob) followed by { kernel_id, file_path, logs }.
• Without file_path: { kernel_id, logs } — 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
kernel_id	Yes	owner/kernel-slug or just kernel-slug (auto-prefixed with your username).

Returns: a zip blob named {kernel-slug}-output.zip followed by { kernel_id, zip_file, files[], logs }.

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 kernel_id supports a short form. If you omit the owner/ prefix, the plugin automatically prepends the authenticated user's username:

my-analysis          →  your-username/my-analysis
alice/her-analysis   →  alice/her-analysis  (unchanged)

Typical workflow

Create Kernel  →  Run Kernel Code  →  Get Kernel Status (poll)  →  Get Kernel Output
                                                                 →  Download Kernel Output (all files as zip)

1. Create Kernel once to get a persistent kernel_id.
2. Run Kernel Code to push new code and trigger execution.
3. Get Kernel Status in a loop until status == "complete".
4. Get Kernel Output to retrieve a specific result file or logs.
5. Download Kernel Output to retrieve all output files packaged as a zip.