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

# MCP Changelog

> Release history, feature additions, and updates for the Labellerr MCP connector and server.

> [!NOTE]
>
> **Latest Release: November 2026** — Launch of the Hosted MCP Connector with 23 comprehensive tools, per-user OAuth sign-in, atomic project setup, secure S3/GCS/Google Drive ingestion, dry-run validations, and Claude-assisted pre-labeling.

***

## November 2026

### Hosted Connector Launch

The Labellerr MCP server is now available as a managed, high-performance **Hosted Connector** with all **23 tools** fully supported and zero local installation required.

* **One URL for Claude**: Add `https://mcp.labellerr.com/mcp` as a custom connector across **Claude.ai**, **Claude Desktop**, and **Claude Code**.
* **Per-User OAuth Sign-in**: Users authenticate with their own Labellerr login in the browser and select their workspace. Tool calls execute under the signed-in user's individual roles and project permissions rather than a single static API key.
* **End-to-End Encryption & Security**: Stored sessions are encrypted; tokens given to Claude are stored only as one-way hashes. Removing the connector immediately revokes access.
* **Structured Claude Outputs**: Tools return concise, structured data models (IDs, names, status flags) and clear next steps, ensuring Claude never outputs API keys, secrets, or raw file contents into the chat.

***

### Features & Capabilities All 23 Tools Available

| Feature | Description | Status |
| :- | :- | :- |
| **`project_setup`** | Creates both the label template and project in one atomic call from a plain label list (`BoundingBox`, `polygon`, `polyline`, `dot`). Pre-validates all inputs before modifying workspace state. | \[!TIP] Available |
| **`connect_cloud_source`** | Directs users to Labellerr's secure web console to connect AWS S3 or Google Cloud Storage buckets. Secrets never pass through conversational chats. | \[!TIP] Available |
| **`get_job_status`** | Universal status checker for asynchronous operations (`dataset_create`, `upload_preannotations`, `export_annotations`). Tracks progress and provides audit error reports. | \[!TIP] Available |
| **`monitor_active_operations` & `monitor_project_progress`** | Real-time tracking of active tasks, overall completion percentages, and review stages. | \[!TIP] Available |
| **`query_project_statistics`** | Granular project analytics and label frequency breakdowns. | \[!TIP] Available |
| **Google Drive Import** | Import images directly from Google Drive links shared as "Anyone with the link". Supports up to 2,500 images and 2.5 GB per import. | \[!TIP] Available |
| **COCO Dry Run Validation** | Pre-upload inspection of COCO JSON files against target projects. Checks exact file matches (detects case discrepancies like `IMG_1.jpg` vs `IMG_1.JPG`), validates categories, and flags out-of-bounds bounding boxes. | \[!TIP] Available |
| **Claude-Drawn Pre-Labels with Review** | Claude inspects sample images using Claude Vision, proposes bounding boxes according to the project's label schema, and presents them for visual confirmation before uploading. | \[!TIP] Available |
| **Resource Discovery** | Added `list_connections`, `list_templates`, and `search_projects` for fast natural-language discovery. | \[!TIP] Available |

***

### Improvements & Fixes

* **Pre-annotation flexibility**: `upload_preannotations` accepts COCO JSON either inline or via public URL (up to 25 MB) and returns a persistent `job_id` immediately upon submission. For larger datasets, `upload_preannotations_async` allows non-blocking asynchronous processing.
* **Data type resilience**: `project_list` and `search_projects` gracefully handle legacy projects without explicit data types without raising exceptions.
* **Cloud ingestion**: `dataset_create` supports folder-level scoping (`path`) within S3/GCS buckets and public Google Drive folder links.

***

## Earlier Releases

### Local MCP Server

The initial release of the Labellerr MCP Server operates locally over `stdio` using Python and Labellerr API credentials (`LABELLERR_API_KEY`, `LABELLERR_API_SECRET`, `LABELLERR_CLIENT_ID`).

* **23 Tools**: Full coverage across Project Management, Dataset Management, Annotation Operations, System Monitoring, and Querying.
* **Direct Local Uploads**: Includes `dataset_upload_files` and `dataset_upload_folder` for streaming local hard drive directories directly into Labellerr.
* **Full IDE Compatibility**: Supports Claude Desktop, Cursor, and Claude Code.
* See [Local MCP Server](https://docs.labellerr.com/mcp/mcp-local-server.md) for full documentation and setup instructions.

***

## Related Documentation

> [card] [Overview & Setup](https://docs.labellerr.com/mcp/mcp-server.md)
>
> Add the hosted connector and sign in.

> [card] [Tools Reference](https://docs.labellerr.com/mcp/mcp-tools-reference.md)
>
> Detailed specifications for all 23 tools.

> [card] [Connect Cloud Storage](https://docs.labellerr.com/mcp/mcp-connect-s3-gcs.md)
>
> Secure AWS S3, GCS, and Google Drive connections.

> [card] [MCP Workflows](https://docs.labellerr.com/mcp/mcp-workflows.md)
>
> End-to-end recipes and automation prompts.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.