Kaggle
Kaggle is a platform for data science and machine learning competitions, offering datasets, notebooks, and a collaborative community.
Verdict
Common use cases
- Download competition datasets for local analysis
- Publish new dataset versions from experiment results
- Pull kernel outputs into team discussions
- Check dataset processing status during uploads
- Locate Kaggle credentials for troubleshooting
Integration
- Vendor
- Kaggle
- Category
- other
- Auth
- API_KEY
- Tools
- 20
- Composio slug
kaggle
Tools
- Create Dataset Version
Tool to create a new dataset version on Kaggle. Use when you have updated files or metadata and need to publish a new version of an existing dataset.
- Dataset Create
Tool to create a new Kaggle dataset with full metadata. Use after uploading files and finalizing metadata. Returns creation status and message.
- Download competition data files
Tool to download competition data files. Use after confirming the competition ID.
- Download kernel output
Tool to download the output of a Kaggle kernel. Use when needing the latest kernel results locally.
- Get Dataset Status
Tool to get the status of a dataset upload or processing job. Use after uploading a dataset to check processing state.
- Get Kaggle Config Directory
Tool to retrieve the directory of the Kaggle API configuration file. Use when you need to locate the directory containing your kaggle.json credentials.
- Get Kaggle Config File
Tool to retrieve the filename of the Kaggle API configuration file. Use when you need to find out where the local Kaggle config file is stored before reading or updating.
- Get Kaggle Config Path
Tool to retrieve local Kaggle API configuration file path. Use when you need to know the location of the Kaggle config before operations.
- Get Kernel Status
Tool to get the status of a Kaggle kernel run. Use after submitting a kernel to monitor its execution state.
- Initialize Kaggle Configuration
Tool to initialize Kaggle API client configuration. Attempts CLI first; if unavailable, it falls back to creating ~/.kaggle/kaggle.json (or $KAGGLE_CONFIG_DIR/kaggle.json).
- Kaggle Dataset Init
Tool to initialize a dataset-metadata.json file in a local folder. Use when preparing a dataset folder before uploading to Kaggle.
- Kaggle Kernel Init
Tool to initialize a kernel-metadata.json file in a local folder. Use when preparing a kernel folder before pushing to Kaggle.
- List Kaggle Configuration Keys
Tool to list local Kaggle API configuration keys. Use when you need to see which configuration options are set without revealing values.
- List Kaggle Dataset Files
Tool to list files in a Kaggle dataset. Use when you need to retrieve paginated file listings by owner and dataset slugs, with optional version and paging controls.
- List Kaggle Datasets
Tool to list Kaggle datasets with filters and pagination. Use after authenticating with Kaggle API key.
- Reset Kaggle Configuration
Tool to reset local Kaggle CLI configuration to defaults. Clears CLI-managed keys ('competition', 'path', 'proxy').
- Set Kaggle Configuration
Tool to set a Kaggle CLI configuration parameter. Use when updating local CLI settings such as default download path or proxy. Ensure Kaggle CLI is installed.
- Submit Competition Entry
Tool to submit an entry to a Kaggle competition. Use when you have already uploaded your file and obtained its blob token.
- Unset Kaggle Configuration
Tool to unset a Kaggle CLI configuration parameter. Use when removing local CLI settings such as default download path or proxy. Ensure Kaggle CLI is installed.
- View Kaggle Configuration
Tool to view local Kaggle API configuration. Use when you need to confirm credentials before API calls.
Setup
Setup guide
- 11. Go to your Kaggle account settings at kaggle.com/settings and scroll to the API section. 2. Click 'Create New API Token' to download a kaggle.json file containing your username and key. 3. Open Switchy, navigate to Settings > Integrations, and find Kaggle in the MCP list. 4. Click 'Connect', then paste the username and key from kaggle.json into the fields provided. 5. Save the connection and return to any Space. 6. Type '@Kaggle get config path' and send — if the MCP replies with a valid file path, the connection works. 7. To use a tool, @mention Kaggle in a message and describe what you need, like '@Kaggle download the Titanic competition data' or '@Kaggle check the status of my dataset upload'. The MCP will invoke the appropriate tool and return results inline.
What teammates see: by default, memories from Kaggle are scoped to the Space (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Spaces have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
Download Competition Data
@Kaggle download the data files for the 'house-prices-advanced-regression-techniques' competition and confirm when complete.Open in a Space →
Create Dataset Version
@Kaggle create a new version of my 'customer-churn-predictions' dataset with the updated CSV file I just generated.Open in a Space →
Check Dataset Status
@Kaggle check the status of the dataset upload I started ten minutes ago and let me know if it's finished processing.Open in a Space →
Download Kernel Output
@Kaggle download the output from my 'feature-engineering-v3' kernel and summarize the key metrics in the logs.Open in a Space →
Locate Config File
@Kaggle get the full path to my kaggle.json config file so I can verify the credentials are correct.Open in a Space →
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@kaggle search for recent datasets about climate change with at least 100 usability votes and show me the top 5 results
Found 5 highly-rated climate datasets: 1. **Global Temperature Anomalies (1880-2023)** — 2,847 usability votes, updated 3 months ago. Contains monthly temperature deviations from baseline across 195 countries. 2. **CO2 Emissions by Country (1960-2022)** — 1,923 votes, updated 6 months ago. Annual emissions data with GDP and population context. 3. **NOAA Weather Station Records** — 1,456 votes, updated 2 weeks ago. Hourly readings from 12,000+ stations worldwide. 4. **Satellite Ice Sheet Measurements** — 892 votes, updated 4 months ago. Greenland and Antarctic mass balance data. 5. **Climate Policy Database** — 654 votes, updated 1 month ago. Legislative actions across 180 nations since 1990. Each entry includes download links and kernel usage examples.
This example shows the MCP's search and filtering capabilities. The actual API returns structured metadata (owner, size, file count, license) that the AI summarizes into readable prose. Requires valid API key with read permissions. Search results depend on Kaggle's live index and may vary by query timing.
@kaggle download the Titanic competition training data to my local datasets folder and confirm the file size
Downloaded Titanic competition files to `/datasets/titanic/`: - `train.csv` (60.9 KB) — 891 passenger records with survival labels - `test.csv` (27.8 KB) — 418 passengers for prediction submission - `gender_submission.csv` (3.2 KB) — example submission format Total download: 92 KB across 3 files. All files verified and ready for local analysis. Note: You'll need to accept the competition rules on Kaggle's website before this download succeeds if you haven't already.
Demonstrates file download with side-effects (writes to local disk). The MCP handles authentication and rule acceptance checks, but users must have previously agreed to competition terms through the web UI. Download paths are configurable. Large competition datasets (10+ GB) may hit rate limits or require resumable downloads not shown here.
@kaggle create a new dataset version for my 'customer-churn-analysis' dataset with the updated CSV I just generated, and write a changelog noting the new feature columns
Created version 3 of `your-username/customer-churn-analysis`: **Changelog:** Added 4 new feature columns (customer_lifetime_value, support_ticket_count, last_interaction_days, product_category) and corrected 127 missing values in the tenure field. Updated schema documentation to reflect new column types. **Status:** Processing (typically completes in 2-5 minutes). The new version will appear in search results once Kaggle finishes validation and indexing. You can check status with the dataset status tool or view it at kaggle.com/datasets/your-username/customer-churn-analysis.
Shows the create-version workflow, which is asynchronous. The MCP submits the request but Kaggle processes uploads server-side. Users need write permissions on the dataset (must be the owner or collaborator). The example assumes files are already staged locally—the MCP doesn't generate CSVs, only uploads them. Version history is public unless the dataset is private.
Use-case deep-dives
When Kaggle MCP makes sense for competition-focused data science teams
A 3-person ML team entering monthly Kaggle competitions needs fast dataset downloads and kernel output retrieval without leaving their workspace. The Kaggle MCP wins here because it automates the download-unzip-organize cycle that otherwise burns 15 minutes per competition. The team can pull competition files, check kernel status, and grab outputs directly in Switchy without context-switching to the Kaggle web UI. The threshold: if your team only enters one competition per quarter, the setup overhead (API key config, learning the 20 tools) outweighs the time saved. But for teams running 4+ competitions a year, this MCP cuts prep friction by half and keeps the entire workflow in one place.
How academic research teams use Kaggle MCP for dataset publishing
A university research lab publishes open datasets on Kaggle as part of paper releases. The Kaggle MCP handles the create-version-upload loop without forcing grad students to learn the Kaggle CLI or web uploader. The lab uses Dataset Create and Create Dataset Version to push new releases directly from Switchy after cleaning data in Python notebooks. The win is speed: versioning a 2GB dataset takes one tool call instead of a 10-step web form. The catch is metadata—if your datasets need complex licensing or detailed provenance fields, you'll still need the web UI for the first publish. For labs pushing 6+ dataset updates a year, this MCP turns a 20-minute admin task into a 2-minute background job.
When Kaggle MCP helps solo consultants pull client benchmark data
A freelance data consultant needs to pull public Kaggle datasets for client benchmarking projects—comparing a retail client's churn model against open competition baselines. The Kaggle MCP works if the consultant already has a Kaggle account and runs 3+ benchmark pulls per month. The Download competition data files tool grabs the exact dataset version used in a past competition, ensuring apples-to-apples comparisons. The downside: Kaggle's API requires manual acceptance of competition rules before first download, so the MCP can't fully automate cold-start projects. If you're pulling datasets once a quarter, just use the Kaggle web UI. But for consultants running recurring benchmark reports, this MCP saves 10 minutes per pull and keeps all analysis artifacts in Switchy's shared workspace.
Frequently asked
What does the Kaggle MCP let me do in Switchy?
It connects your Kaggle account so you can download competition datasets, create and version your own datasets, pull kernel outputs, and manage dataset uploads—all from Switchy's AI workspace. You don't need to leave the chat to grab CSVs or check if your dataset finished processing. It's built for teams running ML experiments who want Kaggle data inside their shared context.
Do I need a Kaggle API key to connect this MCP?
Yes. The MCP uses API key authentication, so you'll need to generate a key from your Kaggle account settings and paste it into Switchy during setup. You don't need admin rights on Kaggle—just a personal account with API access enabled. The key lives in your Switchy workspace and isn't shared with other team members unless you explicitly grant them access to the integration.
Can the Kaggle MCP submit competition entries or run kernels?
No. It can download competition data and pull kernel outputs, but it won't submit predictions or execute new kernels for you. If you need to run code on Kaggle's infrastructure, you still do that in the Kaggle UI or via their CLI. This MCP is read-heavy: it's for pulling datasets and results into Switchy, not for pushing compute jobs back to Kaggle.
How is this different from just using the Kaggle CLI?
The CLI requires you to context-switch out of your chat, remember the right command syntax, and manually paste results back into Switchy. The MCP lets your AI assistant run those commands for you—download a dataset, check upload status, create a new version—without leaving the conversation. It's faster for teams who want Kaggle data in their shared workspace without the terminal friction.
Who on my team should connect the Kaggle MCP?
Whoever owns the Kaggle account you want to pull data from. If your team shares a single Kaggle login for competition datasets, that person connects it. If everyone has their own account, each member can connect their own key. The MCP doesn't count against Switchy plan limits—it's just another integration slot in your workspace.