Skip to main content

Introduction

The Labellerr MCP Server lets AI assistants such as Claude work directly with your Labellerr workspace. Built on the Model Context Protocol (MCP), it turns plain-language requests into Labellerr actions: find datasets, import data from cloud storage or Google Drive, create label templates and projects, upload pre-annotations, track live progress, and export finished annotations.
Example Prompt
Labellerr offers the MCP server in two forms:

Hosted Connector

Recommended: Add Labellerr to Claude with one URL and sign in securely with your Labellerr account.

Local Server

Run the server on your own machine using API credentials when you need local file uploads or Cursor support.

Connect the Hosted Connector

Prerequisites

  • A Labellerr account with access to at least one workspace.
  • One of: Claude.ai (web), Claude Desktop, or Claude Code.
  • Permission to add custom connectors in Claude (on Team and Enterprise plans, a workspace owner may need to allow custom connectors).

Connector URL

Connector URL
Safe to Share: The URL holds no secret or credentials. Every person signs in with their own Labellerr account, so it is completely safe to share this URL across your team.

Add the Connector

1

Open Connectors

Navigate to Settings → Connectors in Claude.ai and select Add custom connector.
2

Add the URL

Name the connector Labellerr and paste https://mcp.labellerr.com/mcp as the URL. Click Add.
3

Sign in

Click Connect. Claude opens the Labellerr sign-in page in a new window. Sign in with your Labellerr account, choose your workspace on the consent page, and click Allow.
4

Enable in Chat

In a new chat, open the tools menu and ensure Labellerr is enabled.

Check the Connection

To verify that the connector is properly configured and active, ask Claude:
Claude will call the health tool, returning:
  • The server status and version
  • Total tool count (23 tools)
  • Your Labellerr client_id
  • The user account you are authenticated as

Sign-in and Permissions

How Sign-in Works

The hosted connector uses standard OAuth 2.0. Claude never sees your password or login credentials:
  1. You sign in on Labellerr’s official login page.
  2. You select your workspace on the consent screen.
  3. Claude receives a scoped connector token valid exclusively with the Labellerr MCP server.
After signing in, every tool call runs with your own user identity. You see the exact same projects and datasets as in the Labellerr web app, and any permission restrictions that apply in the web app are enforced through Claude as well.

What the Connector Stores

  • Encrypted authentication: Your Labellerr sign-in session is stored encrypted, preventing you from having to sign in repeatedly.
  • One-way hashed tokens: Tokens issued to Claude are stored only as secure one-way hashes.
  • Zero data retention: The connector does not store your datasets, files, labels, or exports. All assets reside securely in Labellerr.
  • No secrets in responses: Tool outputs never include API keys, passwords, or cloud credentials.

Signing in Again

Your authentication token remains active for a standard session period. When it expires, Claude displays a Connect or Authenticate prompt for Labellerr. Simply authenticate again; any ongoing operations (imports, uploads, exports) continue running in Labellerr uninterrupted.

Disconnecting

To revoke access, remove the Labellerr connector in Claude (Settings → Connectors, or claude mcp remove labellerr in Claude Code). Removing it immediately revokes the token and clears the stored session. Your active session in the Labellerr web app is not affected.

What You Can Do

Find Your Work

List, search, and inspect projects, datasets, templates, and saved cloud connections using natural language.

Import Cloud & Drive Data

Connect AWS S3 or Google Cloud Storage buckets securely, or import directly from Google Drive links.

One-Step Project Setup

Provide a list of label names and a dataset; Claude builds the template and launches the project together.

Pre-annotations & Exports

Run dry-run validations on COCO files, generate Claude pre-labels with visual review, track project metrics, and export annotations in COCO, JSON, CSV, or PNG.

Example Requests


Feature Status

All 23 features of the Labellerr MCP Server are fully available:

Troubleshooting

  • Verify that the connector is enabled for your current chat (check the tools menu in Claude.ai or Claude Desktop, or run /mcp in Claude Code).
  • If the connector prompts with Connect or Authenticate, sign in again.
  • Start a new chat session after adding the connector.
  • Make sure you use the exact email and password associated with your Labellerr web app account.
  • On the consent screen, verify you select a workspace where you have active membership.
  • If you previously canceled the consent window, remove the connector in Claude and add it again.
  • Your session token has expired. Click the connector prompt in Claude to authenticate again.
  • Long-running operations (imports, uploads, exports) are not lost when a session expires; they continue executing on Labellerr servers.
  • Ask Claude: “Check my Labellerr connection.”
  • The health tool will verify whether the server is reachable and identify the authenticated user.
  • If connectivity fails, contact Labellerr Support with the error timestamp.

Next Steps

Tools Reference

Comprehensive reference for all 23 tools, inputs, outputs, and limits.

Connect S3, GCS & Drive

Learn how to connect cloud storage and Google Drive without sharing credentials in chat.

Workflows & Examples

End-to-end recipes from cloud data to project creation, dry runs, and export.

Local MCP Server

Run the server on your local machine with Python and API keys.

MCP Changelog

Read the latest release notes and updates.