Unitlab AI

official

Create projects, upload data, and automate data annotation directly through your AI assistant with Unitlab MCP.

What can you do with Unitlab AI MCP?

Explore workspace context — Ask your assistant to list projects, inspect ontologies, and summarize assigned annotation tasks via https://api.unitlab.ai/mcp.
Run AI-assisted labeling — Instruct the assistant to detect objects, segment items, or refine labels using your project's classes and save them for review.
Manage dataset releases — Have the assistant check release readiness, identify blockers, create releases, and retrieve exports in supported formats.
Handle annotation workflows — Direct the assistant to claim tasks, save annotations, validate labels, and move work through review stages.
Monitor quality and consensus — Ask the assistant to review consensus results, submit votes, and track validation issues or benchmark scores.

Hosted MCP Server

npx add-mcp 'https://api.unitlab.ai/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Connect withClaudeCodexChatGPTCursorGemini CLIVS Code

One server.Your choice of assistant.

Add Unitlab, sign in, and choose your workspace. Your assistant is ready to use the tools you allow.

Connect Claude Code

Run the command in your project.
Open /mcp in Claude Code and sign in to Unitlab.
Choose your workspace and permissions, then select Allow.

For every project, add --scope user to the command.

Official client guide

claude mcp add --transport http unitlab https://api.unitlab.ai/mcp

Connect Claude

Open Customize → Connectors and add a custom connector.
Name it Unitlab, paste the address, and follow the connection prompts.
Sign in to Unitlab, choose your workspace and permissions, then select Allow.

On Team and Enterprise, an owner adds the connector for the organization before members connect.

Official client guide

https://api.unitlab.ai/mcp

Connect Codex

Run the command with the Codex CLI installed.
Complete the Unitlab sign-in that opens in your browser.
Choose your workspace and permissions, then select Allow.

If sign-in does not open, run codex mcp login unitlab.

Official client guide

codex mcp add unitlab --url https://api.unitlab.ai/mcp

Connect ChatGPT

Open ChatGPT Plugins, select +, then Add custom MCP server.
Add a name and description. Paste the address under Connection and configure OAuth authentication.
Create and install the plugin, completing Unitlab sign-in and consent when prompted.
Use @ to select the plugin in a new chat.

Custom MCP availability depends on your ChatGPT plan and workspace policies.

Official client guide

https://api.unitlab.ai/mcp

Connect Cursor

Add this entry to your Cursor MCP configuration. Preserve any existing servers.
Connect Unitlab in Cursor and complete sign-in when prompted.
Choose your workspace and permissions, then select Allow.

Use the project’s.cursor/mcp.json for a project-specific connection.

Official client guide

{
  "mcpServers": {
    "unitlab": {
      "url": "https://api.unitlab.ai/mcp"
    }
  }
}

Connect Gemini CLI

Run the command with Gemini CLI installed.
Run /mcp auth unitlab inside Gemini CLI.
Sign in to Unitlab, choose your workspace and permissions, then select Allow.

Add --scope user to make the server available across projects.

Official client guide

gemini mcp add --transport http unitlab https://api.unitlab.ai/mcp

Connect VS Code

Run this command with the VS Code command-line launcher installed.
Start the Unitlab MCP server in VS Code and sign in when prompted.
Choose your workspace and permissions, then select Allow.

Use an MCP-enabled chat mode in VS Code. Client and organization policies still apply.

Official client guide

code --add-mcp "{\"name\":\"unitlab\",\"type\":\"http\",\"url\":\"https://api.unitlab.ai/mcp\"}"

Connect your MCP client

Add a remote MCP server using this address.
Use a client that supports Streamable HTTP and OAuth sign-in.
Sign in to Unitlab, choose your workspace and permissions, then select Allow.

For another Unitlab deployment, copy its address from Settings → Connected assistants.

https://api.unitlab.ai/mcp

Prefer to start in Unitlab? Open Settings → Connected assistants → Connect assistant.

Open Unitlab

97 tools for yourdata workflow.

From the first upload to the next dataset release. Your assistant calls these tools within your role and connection permissions.

Explore the Unitlab MCP catalog

Category	What your agent can do	Tools
Getting started	Understand your connection, get practical guidance, and diagnose blockers. get_contextdiagnoseget_help	3
Projects & ontologies	Create projects, define classes and properties, and manage ontology versions. list_ontologiesget_ontologycreate_ontologyupdate_ontologyset_ontology_versionget_instructions Show 4 more tools	10
Data & curation	Upload, import, organize, group, tag, and version your data. list_assetsprepare_uploadcomplete_uploadget_upload_statusattach_datadetach_data Show 10 more tools	16
Content & annotation	Read source content, inspect annotation schemas, save labels, and paint LiDAR points. get_annotation_schemaget_annotationsvalidate_annotationssave_annotationsread_contentpaint_points	6
AI-assisted labeling	Detect, segment, refine, find similar objects, and track supported objects. segment_objectdetect_objectsfind_similar_objectsrefine_objectstrack_object	5
Models & examples	Discover models, manage reference examples, and follow labeling runs. list_modelsget_model_runstart_model_runcancel_model_runget_referencessave_reference_examples Show 3 more tools	9
Tasks & workflows	Claim and assign tasks, configure workflow stages, and move work forward. list_tasksget_taskget_next_taskget_groupclaim_taskassign_tasks Show 5 more tools	11
Quality & consensus	Review Consensus results, validation issues, and benchmark scores. claim_consensus_taskget_consensus_tasksubmit_consensus_voteget_consensus_reviewset_consensus_approvalfinish_consensus_review Show 5 more tools	11
Keep comments, issues, and notifications connected to the work. list_commentsadd_commentresolve_commentlist_issueslist_notificationsmark_notifications_read	6
Releases & exports	Create, monitor, download, clone, and manage dataset releases. list_releasesget_releaseretry_releasecreate_releaseget_release_downloadclone_release Show 2 more tools	8
Members & roles	Manage workspace membership and roles when authorized. list_membersupdate_memberlist_rolescreate_roleupdate_role	5
Administration & statistics	Manage integrations and API keys, inspect audit events, and read statistics. create_api_keyupdate_api_keylist_cloud_storagescreate_cloud_storageupdate_cloud_storagelist_audit_events Show 1 more tools	7

No matching tools. Try a different name or category.

Need a hand? get_help explains workflows and tools. diagnose checks what is blocking your work.

A prompt for every stageof your data workflow.

Copy an example to get started. Your assistant works with the same projects, labels, and review stages your team uses in Unitlab.

Understand your workspace

WorkspaceProjectsTasks

Find your projects, inspect the context, and see the annotation work assigned to you.

Explore data curation

Try this prompt[

Copy prompt

](#)

Show my Unitlab projects and summarize the annotation work assigned to me.

Read access

Annotate and review

OntologyAI labelsReview

Create labels with your classes, run supported AI tools, and move work through annotation and review.

Explore annotation

Try this prompt[

Copy prompt

](#)

In my image project, use the Car class to detect visible cars on the next task. Save the labels and show them for review.

Write access · AI usage

Prepare your next release

ReadinessReleaseExport

Inspect blockers, create dataset releases, and retrieve exports in the formats your project supports.

Explore dataset management

Try this prompt[

Copy prompt

](#)

Check whether my project is ready for a release. Show any blockers and the supported export formats before creating it.

Read first · Write to create

Your workspace.Your permissions.

Connect as yourself. Choose one workspace and decide what your assistant can do.

Manage in Unitlab

Choose read or write access

Explore with Read only, or allow changes with Read and write. Your role and workspace access still apply.

Choose the permissions

Customize access during sign-in or edit your own connection later in Connected assistants.

Stay in control

Disconnect in Unitlab at any time. Reconnect to change the workspace or access level.

Some actions save changes or start jobs immediately. Some AI operations and model stages use your workspace’s AI quota.

Frequently askedquestions.

Read the MCP documentation

What is the Unitlab MCP server?

The Unitlab MCP server connects compatible AI assistants to your Unitlab workspace through Model Context Protocol. It exposes tools for data management, annotation, AI-assisted labeling, tasks, review, workflows, and dataset releases. Your assistant calls those tools using the workspace and permissions you authorize. Read the MCP overview →

Which AI assistants can I connect?

Unitlab provides setup instructions for Claude Code, Claude web and desktop, Codex, ChatGPT, Cursor, Gemini CLI, and VS Code. Other clients need support for remote MCP over Streamable HTTP and OAuth. The available interface and connection options depend on the client version, plan, and workspace policy. Connect your assistant →

How do I connect, and do I need an API key?

Add https://api.unitlab.ai/mcp in your assistant, then complete Unitlab sign-in. Select a workspace, choose Read only or Read and write, customize the permissions if needed, and click Allow. This OAuth connection does not require you to create or paste an API key. Unitlab also provides setup instructions under Settings → Connected assistants. Follow the quickstart →

Can I upload a folder and label the images in one conversation?

Yes. A compatible assistant can create a project, set up its classes, and upload the images in a folder. Direct file upload requires the assistant to send file bytes through HTTP. It can coordinate detection for named classes across images, or configure a model workflow for batch labeling. Results depend on the data, classes, model, and quota, and should be reviewed in Unitlab. See the folder upload workflow →

Which data types can the assistant work with?

The MCP tools support image, video, LiDAR/3D, medical, pathology, geospatial, audio, text, document, HTML, sensor, and tabular data. Grouped multimodal assets are supported too. Reading, annotation types, AI operations, and exports vary by modality. The assistant can request the item’s annotation schema and built-in data-type guidance before making changes. Explore the tool reference →

Can the assistant change data or start AI jobs?

Yes, with a Read and write connection, the required role permissions, and applicable workspace access. Tools can save annotations, modify workflows, and start model runs. Some AI-assisted operations and model stages consume the workspace’s AI quota. Choose Read only if you want exploration without writes; approval behavior inside the assistant depends on that client. Understand access and AI usage →

How do permissions and disconnecting work?

Each connection acts as the member who authorized it in one selected workspace. Your role and any permissions you turn off limit the available tools. In Settings → Connected assistants, use Edit permissions for your own connection or Disconnect to revoke access. To select another workspace or read/write access level, reconnect and approve the new settings. Manage permissions and connections →

Does MCP replace the annotation editor or human review?

MCP adds another way to work with Unitlab. Your team can continue reviewing and refining the same data in Unitlab’s annotation interface. The assistant can read annotated previews, check validation problems, and follow task and review stages. Generated labels still need the review appropriate to your project. Follow annotation and review workflows →

What if a tool or task is unavailable?

Ask the assistant to check its connection context and call Unitlab’s built-in help. The get_help tool explains workflows, tools, data types, and errors; diagnose checks blockers on a project, item, or task. A missing permission, unavailable workspace access, unsupported operation, or exhausted quota can limit what a connection can do. Open the troubleshooting guide →