Introduction
With the Labellerr MCP Connector, you can import data directly from your Amazon S3 buckets, Google Cloud Storage buckets, or shared Google Drive links using natural language:Example Prompt
How It Works
Connection Flow
1
Ask Claude to connect storage
Claude calls
connect_cloud_source and returns a direct link to the Create Dataset screen of the Labellerr web console, along with step-by-step instructions.2
Save connection in Labellerr
In the Labellerr web app, you paste your credentials, click Test Connection, and save the connection. Labellerr validates your credentials before saving.
3
Claude identifies the connection
Once you notify Claude that the connection is saved, Claude invokes
list_connections. The newest connection appears at the top; Claude sees only its ID and human-friendly name.4
Claude triggers the import
You specify the target folder within the bucket (e.g.
retail/shelves/), and Claude invokes dataset_create. Labellerr launches an asynchronous import job and returns a job_id.5
Claude notifies you upon completion
Claude periodically checks
get_job_status (or dataset_get) until the dataset enters ready status. You can immediately launch an annotation project on the new dataset.Already have saved connections? You can skip setup entirely. Just ask: “Which S3 connections do I have?” and tell Claude to import from an existing connection.
Before You Start: Cloud Permissions
Ensure your bucket credentials provide the necessary read permissions for Labellerr.Amazon S3
Create an IAM User or IAM Role with the following permissions on your target bucket:
For detailed AWS IAM setup instructions, refer to our AWS S3 Connection Guide.
Google Cloud Storage (GCS)
Generate a Service Account JSON key file with the following IAM roles or permissions:
For detailed Google Cloud IAM setup, refer to our GCS Connection Guide.
Step 1: Save a Connection in Labellerr
Ask Claude:connect_cloud_source and provides a direct URL and guided steps:
- Amazon S3
- Google Cloud Storage
- Open the link provided by Claude and sign in to Labellerr if requested.
- Enter any placeholder dataset name, choose your data type (e.g., Image), and click Next.
- Select Amazon S3, then click Create new connection.
- Enter your Access Key ID, Secret Access Key, and the S3 path (
s3://your-bucket-name/path/to/data/). Click Test Connection. - When the test succeeds, enter a memorable connection name (e.g.
retail-prod-s3) and click Save. - Click Cancel to exit the Create Dataset wizard (the connection remains securely saved in your workspace).
- Return to Claude and say: “Done, the connection is saved.”
Step 2: Import a Dataset
Claude callslist_connections, retrieves the newly saved connection, and prompts for the subfolder to import:
Example Prompt
dataset_create with:
The dataset is initialized in
processing status. Labellerr scans the target path and imports all files matching the specified data type (for images: .jpg, .jpeg, .png, .tiff).
Step 3: Monitor Import Status
Imports run on Labellerr cloud servers, so they proceed even if you close Claude or shut down your computer. You can check progress at any time:Prompt
get_job_status (or dataset_get):
Google Drive Import
You can import datasets directly from Google Drive without setting up IAM policies or cloud access keys:Available: Google Drive links can be imported directly through Claude!
How to Use Google Drive Import
- In Google Drive, right-click the folder or file you wish to import.
- Select Share → Share, and set General Access to “Anyone with the link” (Viewer permission).
- Copy the link and share it directly with Claude:
Example Prompt
Specifications and Limits
- Maximum Files: Up to 2,500 images per import.
- Maximum Size: Up to 2.5 GB total dataset size.
- Supported Formats: JPG, JPEG, PNG, TIFF.
- Error Handling: Links not set to “Anyone with the link” fail immediately with an explicit permission error. Claude reports any skipped or unreadable files alongside the reason.
Security Architecture
Where do cloud credentials reside?
Where do cloud credentials reside?
Credentials never touch Claude, the conversational prompt, or the MCP connector layer. Access keys and service accounts are transmitted over HTTPS directly to Labellerr’s backend and stored encrypted in accordance with SOC 2 compliance standards.
What can Claude see?
What can Claude see?
Claude can only see metadata returned by
list_connections: connection ID, connection label/name, provider (s3 or gcs), connection type, and creation date. Folder paths passed in chat are visible to Claude, so avoid using proprietary identifiers in bucket path strings if needed.Who can use a saved connection?
Who can use a saved connection?
Claude inherits your authenticated Labellerr workspace identity. It cannot access connections belonging to other workspaces or projects where you lack permissions.
How to remove a connection
How to remove a connection
The MCP connector intentionally does not expose deletion tools to prevent accidental loss. To delete a connection, remove it via the Labellerr web console under Settings → Data Connections or use the Labellerr Python SDK.
Troubleshooting
Connection not found
Connection not found
- Ensure you saved the connection under the exact same workspace you selected when signing in to Claude.
- Ask Claude: “List my S3 connections” or “List my GCS connections” to see all registered connections.
Test Connection fails in Labellerr web app
Test Connection fails in Labellerr web app
- AWS S3: Verify that the Access Key and Secret Key are active and possess
s3:GetObjectands3:ListBucketpermissions. Ensure the bucket path usess3://bucket-name/folder/format with a trailing slash. - Google Cloud Storage: Ensure you uploaded the correct JSON service-account key file and that the service account has
Storage Object Viewerprivileges.
Dataset has 0 or fewer files than expected
Dataset has 0 or fewer files than expected
- S3 and GCS paths are case-sensitive. Verify the exact spelling and capitalization of bucket folder names.
- Files with unsupported extensions are skipped automatically. Ensure your images are
.jpg,.jpeg,.png, or.tiff.
Import failed
Import failed
- Ask Claude: “What was the error for the Shelf Images import?”
- Claude reads the
errorattribute fromget_job_status. Correct the root cause (typically IAM permissions or incorrect bucket name) and re-trigger the import.
Related Documentation
MCP Tools Reference
Inspect inputs and outputs for
dataset_create, connect_cloud_source, and list_connections.MCP Workflows
See end-to-end workflows chaining cloud imports with project creation and exports.
AWS S3 Setup Guide
Step-by-step AWS IAM user and bucket policy creation.
GCS Setup Guide
Step-by-step Google Cloud service account key configuration.

