Breadcrumbs

Using the Harbr MCP


Connecting your AI assistant to the Harbr platform lets you ask for data in plain language instead of navigating the platform yourself. Once connected, an assistant such as ChatGPT or Claude can carry out the workflows enabled for your environment, all from a normal chat conversation, with no coding required.

Your assistant always acts as you. It signs in with your normal platform account and can only ever see or do what your account allows: Products, Plans, or Assets you cannot access in the platform never appear and cannot be discussed.

Note: Your platform administrator must have enabled the AI assistant connection and will provide you with the connection URL, which ends in /mcp. You sign in with your usual platform account, you never copy or paste any keys or tokens. If you are the administrator setting this up, see Setting Up and Configuring the MCP Server.

Connect Your AI Assistant

You only need to do this once. After connecting, the platform is available in every conversation with your assistant. The steps below cover ChatGPT and Claude as examples, but any AI assistant that supports connecting to custom MCP servers can be paired with Harbr in the same way, using the same connection URL.

ChatGPT

Available on ChatGPT Plus, Pro, Business, Enterprise, and Edu, on the web. Adding a custom connector requires Developer mode.

  1. Go to Settings > Apps & Connectors > Advanced settings and enable Developer mode.

  2. Go to Settings > Apps & Connectors and click Create.

  3. Complete the setup:\n

    • Name: the name your administrator recommends for the connector

    • URL: your connection URL (ending in /mcp)

    • Authentication: OAuth

  4. Confirm that you trust the application, then click Create.

  5. Sign in with your platform account in the browser window that opens.

  6. In a chat, click + near the message box to bring the platform into the conversation.

Note: On Business, Enterprise, and Edu workspaces, your ChatGPT administrator may need to allow custom connectors before this option appears.

Claude

Available on Free, Pro, Max, Team, and Enterprise plans, on claude.ai and the Claude desktop and mobile apps.

On individual plans:

  1. Go to Customize > Connectors.

  2. Click +, then Add custom connector.

  3. Enter your connection URL (ending in /mcp) and click Add.

  4. Sign in with your platform account in the browser window that opens.

On Team and Enterprise plans, an Owner must first add the connector for the organization under Organization settings > Connectors. Each member then goes to Customize > Connectors, finds the connector, clicks Connect, and signs in.

Note: On the Free plan, you are limited to one custom connector.

Other AI assistants

Harbr's MCP connection follows the standard Model Context Protocol, so any assistant that supports adding a custom MCP server, such as Claude Code or other MCP-compatible tools, can connect using the same connection URL and the same OAuth sign-in with your platform account. Check that assistant's own documentation for where to add a custom connector or MCP server.

Discovery

Ask your assistant for data in your own words, by topic, by what a Product contains, or by what you want to do with it. It searches the catalogue and returns matching Products with links to their Product pages.

Example: "Find Products with retail footfall data", "Which Products can I export?", "Show me Products containing visualisations".

Your assistant can now act on your behalf for several workflows, not just discover data. For each product, it will tell you in plain language what you can and cannot do with it. In this release that covers subscribing, exporting, and creating a data share. Which workflows are available to you depends on your Harbr environment's configuration, your Harbr account team can adjust this on request.

If there is a relevant product on the exchange then it will be surfaced to the user.

Before requesting access, you can look inside a Product:

  • Ask what a Product contains to see its Assets.

  • Ask about a specific Asset to see its detail, you will be shown details including its schema and column names (capped at 100 columns per Asset).

  • Ask what you can do with a Product, your assistant can tell you which actions your account permits, based on your existing permissions in the platform.

Every Product your assistant shows you, whether from a search or a direct lookup, comes with a link to its Product page as standard, so you can always continue in the platform directly.


Subscribing to a product

Ask your assistant to subscribe you to a product. If the plan is self-serve, your assistant completes it immediately, showing you the terms and getting your agreement first. If the plan needs approval, your assistant submits the request and can check its status for you at any time. If you already hold a subscription that is on hold pending terms acceptance, your assistant can show you the terms and activate it.

Viewing and accepting terms is part of subscribing, there is no separate step.

Example: "Subscribe me to this product", "Has my request for this plan been approved yet?", "Activate my pending subscription to this product."

Note: payment and managed plans cannot be completed by your assistant. It will direct you to the platform or your sales contact instead.

Creating a data share

Ask your assistant to create a data share of a product or asset over Databricks. Two options are available:

  • Open share: produces a downloadable, token-based share you can hand to any recipient using Databricks open sharing. The access token cannot be rotated by your assistant and is valid for one hour, after that, use the platform to issue a new one.

  • Databricks-to-Databricks share: shares directly into the recipient's own Unity Catalog. You will need the recipient's Unity Catalog metastore ID.

Open share works for tabular data only. If you need to share a file asset, use a Databricks-to-Databricks share instead, or export it.

The open share credential file cannot be re-downloaded. Store it securely, if it is lost or compromised, your platform administrator will need to rotate the share's token rather than reissue the file.

Databricks Delta Sharing has a limit of 100 assets per product. Above this limit, creating a share will fail, if you hit this, export the product if your plan allows it.

An asset does not need to be part of a product to be shared this way, a standalone asset can be data shared directly.

Example: "Create a data share of this product for a recipient using their Unity Catalog metastore ID [id]", "Set up an open share of this asset."

Exporting data

Ask your assistant to export a product. It can set the destination (for example Download, S3, GCS, or SFTP), the export's name and description, whether it repeats on a refresh schedule, and email notifications.

Export always works at the product level, it exports every asset of the chosen type within that product. Your assistant cannot export a standalone asset that isn't part of a product, pick out individual assets within a product, add filters to a single export, delete an export, or rename or re-describe one after it is created. For any of these, use the export form in the platform, your assistant will give you the link.

Example: "Export this product to my S3 destination", "Set up a weekly export of this product with email notifications."

Troubleshooting

  • A tool or workflow you expect is not offered: your assistant may be using a cached list of what it can do, particularly after a Harbr release adds new capability. Disconnect and reconnect the Harbr connector in your AI assistant (for example, in Claude's Connectors settings) to refresh the list.

  • Your assistant cannot see the platform: check the connector is enabled in the conversation and that you completed the sign-in step. If it still fails, confirm the connection URL with your administrator, it must end exactly in /mcp.

  • Sign-in shows an account or tenant error (for example, a message saying your account does not exist in a tenant and needs to be added as an external user): this is not the same as a normal login problem. It means your account has not yet been provisioned for this connection by your identity provider, a step separate from your everyday platform login. Contact your platform administrator to have you added, then try connecting again.

  • "Not authorized": you are connected, but your account does not have access to that specific Product, Plan, or Asset, the same permissions apply as when you use the platform directly. Contact your administrator if you believe you should have access.

  • No sign-in window appears, or the connector will not connect: double-check the connection URL is exact and ends in /mcp, and that your plan supports custom connectors.

  • You cannot find where to add a connector: in ChatGPT, make sure Developer mode is on (and on Business, Enterprise, and Edu workspaces, check that your admin has allowed custom connectors); on Claude Team or Enterprise plans, an Owner must add the connector before members can connect. For other assistants, check that assistant's own documentation for adding a custom MCP server.

See also: Setting Up and Configuring the MCP Server, if you are an Administrator looking to configure and enable this connection.