> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/api-sdk-cli/mcp/overview.md).

# Unitlab MCP

Connect your AI assistant to Unitlab with Model Context Protocol to organize data, annotate across modalities, automate workflows, and export datasets.

Unitlab MCP connects your AI assistant to your Unitlab workspace through the **Model Context Protocol (MCP)**. Ask your assistant to inspect projects, upload data, create labels, manage annotation workflows, review quality, or export a dataset. The assistant uses Unitlab tools within the workspace and permissions you approve.

The connection works with compatible clients such as Claude, Claude Code, ChatGPT, Codex, Cursor, Gemini CLI, and VS Code. Your assistant remains the conversational interface; Unitlab stores your projects, annotations, workflow state, and releases.

## Connect to Unitlab

The hosted Unitlab MCP server address is:

```
https://api.unitlab.ai/mcp
```

| Setting        | Value                                                        |
| -------------- | ------------------------------------------------------------ |
| Transport      | Remote MCP over Streamable HTTP                              |
| Authentication | OAuth sign-in with your Unitlab account                      |
| Workspace      | One workspace selected during authorization                  |
| Access         | Read only or Read and write, with optional permission limits |

You do not need to create or paste a Unitlab API key for this connection. For a dedicated Unitlab deployment, copy the server address from that deployment's **Settings → Connected assistants**.

Start with [Connect your AI assistant](https://docs.unitlab.ai/api-sdk-cli/mcp/connect-your-assistant), then follow the [quickstart](https://docs.unitlab.ai/api-sdk-cli/mcp/quickstart).

## What you can do

| Goal                   | Examples                                                                                                                     |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| Explore your workspace | Find a project, inspect its classes and instructions, read items, and check progress.                                        |
| Prepare data           | Create projects, upload files, import from connected cloud storage, organize library assets, and group related files.        |
| Annotate               | Read the content and annotation schema, create or revise labels, and validate saved results.                                 |
| Use AI labeling        | Detect a named object class, segment from a box or clicks, find similar objects, refine shapes, and track supported objects. |
| Automate workflows     | Configure stages, assign work, run a model before human annotation, and monitor processing.                                  |
| Review quality         | Work through review and Consensus, inspect benchmark results, and discuss items with comments.                               |
| Export datasets        | Create versioned releases, monitor export progress, and retrieve a download link.                                            |

The current catalog contains **101 tools**. Your assistant may show a smaller set because tool visibility and execution depend on your role and connection permissions. Browse the [tool reference](https://docs.unitlab.ai/api-sdk-cli/mcp/tools).

## Example: from an image folder to a labeled project

After connecting an assistant that can access and transfer your files, you can ask:

> Create a Road scenes project from this folder. Upload every image and label cars, people, and bicycles across the dataset. Leave the results for my review.

The assistant can check its access, create the project, add classes, upload the image batch, and coordinate labeling across the images. You then inspect the saved labels in Unitlab and continue through your review workflow.

Folder upload requires a client that can read the files and send their bytes. If your assistant cannot transfer files, upload them in the Unitlab web app or ask it to use assets already in your workspace library. AI detection searches for the requested classes; results should be reviewed for missed objects, incorrect detections, and missing properties.

The [quickstart](https://docs.unitlab.ai/api-sdk-cli/mcp/quickstart) explains this process in detail. For automatic pre-labeling whenever data enters a Model stage, see [Workflows and automation](https://docs.unitlab.ai/api-sdk-cli/mcp/workflows#batch-auto-labeling-with-a-model-stage).

## Multimodal annotation

Unitlab MCP supports workflows across these data families:

* Images and video
* Text, HTML, and tabular data
* Audio and time series
* Medical volumes and PDF documents
* Geospatial and pathology imagery
* LiDAR and point-cloud scenes

Reading and annotation operations are tailored to the item. AI tools have narrower modality support. For example, image detection does not operate on a point cloud, while LiDAR provides 3D geometry, camera views, and point painting. Your assistant can request the current schema with `get_annotation_schema` and a modality guide with `get_help`.

See [Multimodal data and coordinates](https://docs.unitlab.ai/api-sdk-cli/mcp/workflows#multimodal-data-and-coordinates) for practical differences.

## Access stays under your control

Each connection acts as the member who authorized it, inside the selected workspace. Unitlab checks your current role, the connection's read/write level, and any permission limits on every call. You can edit your connection's permissions or disconnect it in **Settings → Connected assistants**.

Read [Authentication and permissions](https://docs.unitlab.ai/api-sdk-cli/mcp/authentication-and-permissions) before connecting an assistant that will change data.

## Frequently asked questions

### Can an assistant create a project and label a whole image folder?

Yes, when the client can access and transfer the files and your connection has the required permissions. The assistant can create the project, upload the batch, define the requested classes, and coordinate labeling across the images. Review the results before using them. Follow the [image-folder quickstart](https://docs.unitlab.ai/api-sdk-cli/mcp/quickstart).

### Do I need an API key?

No. The standard MCP connection uses OAuth sign-in. Choose the workspace and read/write access during authorization. API keys remain a separate option for SDK and HTTP integrations. See [Authentication and permissions](https://docs.unitlab.ai/api-sdk-cli/mcp/authentication-and-permissions).

### Which annotation modalities can I use?

The toolset covers twelve data families, including images, video, medical, geospatial, pathology, text, audio, and LiDAR. Reading, annotation, and AI capabilities differ by modality. See [Multimodal data and coordinates](https://docs.unitlab.ai/api-sdk-cli/mcp/workflows#multimodal-data-and-coordinates).

For more answers, see [MCP troubleshooting and FAQs](https://docs.unitlab.ai/api-sdk-cli/mcp/troubleshooting#frequently-asked-questions).

## Documentation

| Guide                                                                                                    | What it covers                                                                   |
| -------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- |
| [Quickstart](https://docs.unitlab.ai/api-sdk-cli/mcp/quickstart)                                         | Connect, inspect a project, upload an image folder, and create the first labels. |
| [Connect your AI assistant](https://docs.unitlab.ai/api-sdk-cli/mcp/connect-your-assistant)              | Setup instructions for supported clients and dedicated deployments.              |
| [Authentication and permissions](https://docs.unitlab.ai/api-sdk-cli/mcp/authentication-and-permissions) | OAuth, workspace scope, roles, permissions, and connection management.           |
| [MCP tool reference](https://docs.unitlab.ai/api-sdk-cli/mcp/tools)                                      | Every tool, organized by task.                                                   |
| [Workflows and automation](https://docs.unitlab.ai/api-sdk-cli/mcp/workflows)                            | Uploads, multimodal annotation, AI labeling, Model stages, review, and exports.  |
| [Troubleshooting](https://docs.unitlab.ai/api-sdk-cli/mcp/troubleshooting)                               | Connection, access, upload, annotation, model, and export issues.                |

Related product guides include [Data curation](https://docs.unitlab.ai/documentation/data/data-curation), [Model stages](https://docs.unitlab.ai/documentation/workflows/model-stages), and [Multimodal annotation](https://docs.unitlab.ai/documentation/multimodal-annotations/multimodal-overview).

For the product overview and example conversation, visit [Unitlab MCP](https://unitlab.ai/en/mcp). For other integration options, see [API, SDK, and CLI](https://docs.unitlab.ai/api-sdk-cli).
