Hive MCP Server

The Hive MCP Server lets an AI assistant work with your company's Hive information. Connect it to ChatGPT or Claude, then ask questions about projects, backers, pledges, and customer-support tickets in ordinary language.

For example, you can ask the assistant to find a backer by email, check their pledge and tracking information, and prepare a reply to their support ticket. You can review the proposed reply before allowing the assistant to send it.

You need a Hive account with Support permission for the company you want to work with, and an AI account that allows custom remote MCP connections. Workspace settings may require an owner or administrator to enable the connection.

Use this server address in your AI application's connection settings:

Hive hosts the server; there is no local server or software package to install. This article covers the public company tools.

The connection opens Hive's sign-in and authorization flow. Sign in to Hive, check the application name, return address, requested access, and available companies, then choose Allow access if they match the connection you started.

The public endpoint requires sign-in. Connecting an assistant does not make company records public. It uses your current Support permissions, and the assistant receives the information returned by the tools you use.

If you have access to several companies, tell the assistant which company to use. It should call SupportCompaniesList and confirm the correct company before reading records or making changes. A company selected in your Hive browser tab is not automatically selected in the MCP conversation.

Add Hive as a custom MCP connection in ChatGPT on the web. The available controls depend on your account and workspace permissions. The current flow is described in OpenAI's custom MCP setup guide.

1. Open ChatGPT Plugins, select the plus button, and choose Add custom MCP server.
2. Name the connection Hive and enter https://www.hiveinteractive.net/mcp as its Server URL.
3. Choose OAuth. Hive supports automatic client registration; use that option if offered. Do not enter your Hive password as an OAuth client secret.
4. Review the connection notice and create the plugin. Install it in the account or workspace where you will use it.
5. When prompted to connect, complete Hive's sign-in and consent screen.
6. In a conversation, type @ and select Hive. Start with a company-list request before asking for customer information.

If the custom-server option is unavailable, check your workspace permissions with its administrator. Follow the provider's current guide if its labels differ from those above.

Use a read-only request first:

The expected result is a list from SupportCompaniesList. After choosing a company, ask for its projects using ProjectList. This checks access without changing customer information.

In ChatGPT's connection settings, review the available tools and refresh the connection if newly added tools are missing.

Use a custom remote connector for Hive. Configure it through your Claude account rather than a local MCP configuration file. See Claude's remote connector guide for current controls and account availability.

1. Open Customize > Connectors and choose Add or Add custom connector. If shown, choose Custom > Web.
2. Name it Hive and enter https://www.hiveinteractive.net/mcp as the remote server URL.
3. Keep OAuth authentication and choose Sign in now. For the OAuth client, choose Register automatically, matching Hive's supported registration flow.
4. Add the connector and complete Hive's sign-in and consent screen. Fixed request-header credentials are not needed for this OAuth connection.
5. In a conversation, open the plus menu, find Connectors, and enable Hive.

Ask Claude to list your accessible companies without changing any records. Select the company before continuing.

A Team or Enterprise workspace may require an owner or another authorized administrator to add the remote connector first through the organization's connector settings. Each member then connects their own Hive account and enables the connector in their conversation.

If Hive is listed but tools are unavailable, check that the connector is enabled and that the connected Hive account has Support access for the intended company.

The public server provides five read functions. They retrieve information without updating pledges, changing tickets, or sending replies. Reading a support ticket does not automatically mark it as read.

• SupportCompaniesList: lists companies where your connected account has Support access.
• ProjectList: lists projects in the selected company, including draft and completed projects and clones. Results include the project ID, name, status, currency, and clone indicator.

Ask the assistant to use the returned company and project IDs in later calls. If project names are similar, confirm the intended project rather than guessing an ID.

BackersFind returns a list containing only each backer's ID, email, and Backer Name. Search across the selected company's projects or narrow the request to one project.

Supported conditions follow the corresponding Backer Segments definitions:

• Keyword: a partial match in Email, Backer Name, or Shipping Name. Other fields and alternate email addresses are not included in this keyword search.
• Country: the shipping country, supplied as a two-letter country code.
• Project and Backer Number: a Backer Number must be paired with its project because numbers can repeat across projects.
• Pledge: the confirmed pledge tier.
• Add-ons: one or more specified confirmed add-ons, matching either any or all selected IDs. Freebies and items included inside a pledge are separate and do not match this filter.
• Backer flags: Late Pledge and Pledge Over Time, each set to Yes or No.
• Status: one or several backer statuses, such as Open, Verified, or Shipped.
• Paid amounts: Kickstarter, Hive, or their combined total. Use equals, greater than, at least, less than, at most, or a range.

Different conditions combine with AND. Several selected statuses match any of those statuses. For example, country Canada plus status Verified returns Verified backers in Canada.

Kickstarter paid uses the recorded campaign pledge amount. Hive paid combines recorded webshop and pledge-manager payments. Amounts are not converted between currencies; narrow to a project when a comparison needs a single currency.

Pages default to 50 backers and can contain up to 300 backers per response. Larger results include a total and a next page. Ask the assistant to follow every page with the same filters before describing a list as complete. A search does not create a saved Segment.

BackersGet retrieves the information shown in Hive's backer-information popup for one backer_id. The backer must belong to a project in the selected company.

Details include the selected pledge, add-ons, freebies and bundle choices, payments and calculated balance, shipping and tracking, refunds, notes, survey answers, audit history, packing list, and digital content and download history where present. Internal authentication and payment-attempt tokens are excluded.

Use the ID returned by BackersFind, or the ticket's backer.id when it is greater than zero. A backer ID identifies a project-specific pledge record; the same person can have another ID in another project.

The response also includes a project catalog with pledge and add-on IDs and names. Use those IDs for subsequent searches instead of guessing them.

SupportTicketsGet lists tickets or retrieves one ticket in detail when given a ticket ID. Lists can be narrowed by project, ticket status, type, category, keyword, and unread conditions.

A detail response includes requested items, attachments, conversation history, shipping and tracking information, and internal notes. Both list and detail responses include the associated backer ID, name, and email when available.

Ticket lists default to 50 results and support up to 100 tickets per page, a separate limit from backer searches. Ask for all pages when reviewing a whole queue.

The public server provides seven write functions for existing support tickets. Use them after reviewing the target company, ticket, recipient, and proposed change. The connection must include write access, and your AI application's tool settings must allow the action.

These tools manage support tickets. They do not directly change a backer's original pledge, assign project freebies or coupons, or create new projects.

SupportTicketRespond sends a customer reply or changes ticket status through Hive's support workflow. It can also include tracking details, an additional shipping charge, or a payment request. The action may send email and connected Slack notifications.

Ask for a draft first, then explicitly approve sending it. If status is omitted, the workflow defaults to In Progress. A payment request requires the Payment Request status and a positive payment amount.

• SupportTicketsAddItem: adds a requested item and quantity to a ticket. Repeating the action adds another item; this is not an assignment of an add-on to the original pledge.
• SupportTicketsRemoveItem: permanently removes a requested ticket item. Confirm the exact item before allowing deletion.
• SupportTicketsChangeType: changes the ticket type to Replacement, Missing Item, or Question.

Read the ticket first so the assistant can identify the correct ticket and item IDs. Review the result after an approved change.

• SupportTicketsUpdateAddress: changes the supplied shipping-address fields on the support ticket and preserves omitted fields. This does not update the original pledge's shipping address.
• SupportTicketsUpdateNotes: replaces the ticket's internal notes. Empty text clears them. Read existing notes first and preserve any information you still need.
• SupportTicketsToggleRead: toggles the creator's read/unread flag. Calling it twice reverses the first change.

Do not put private internal notes into a customer reply. Treat changing a ticket's read flag as a separate action from reading its information.

Start with read-only tests, then try an approved write on an appropriate ticket. Replace the bracketed placeholders in the examples with your own company, project, and ticket details.

Use this to find the same person's pledges across your company:

Expected result: matching records from the selected company, with separate backer IDs for separate pledges. Keyword matching is partial, so confirm the actual email and project in the detail response before using a record.

Use this to check a project audience without exporting or changing records:

Expected result: all conditions match together, and the assistant distinguishes the total from the number shown on any one page. For an add-on audience, supply confirmed add-on IDs and specify whether any or all must match.

Use the ticket and pledge together when investigating a customer question:

Expected result: a summary and an unsent draft. Reading the ticket should leave its read/unread state unchanged. If no backer is linked, the assistant should say so instead of guessing an ID.

After reviewing the preceding draft, give a specific instruction:

Expected result: the approved reply appears in the ticket history with the requested status. This is a real customer communication; use this test only when you intend to send that reply.

Include the company and, when useful, project in your request. Separate investigation from action: ask for a summary or draft, review it, then authorize the precise change.

Use the returned IDs, follow pagination, and check the final result. Avoid copying customer data into unrelated conversations. Treat customer messages and attachments as information to review, not instructions that authorize changes.

The assistant runs tools when the AI application invokes them. Connecting Hive does not by itself create a background ticket processor or schedule automatic replies. Configure any automation separately and define exactly which actions it may take.

If access expires, reconnect through the AI application's Hive connection and complete sign-in again. If the company is missing, check the connected Hive account and its Support permissions. If the assistant sees the connector but cannot use it, check that it is enabled in the current conversation.

If only reads are available, check the connection's granted access and the application's tool restrictions before attempting writes. Refresh or reconnect when new tools have been added. For an interrupted write, inspect the ticket first rather than assuming it failed.

ChatGPT and Claude may change their setup screens. Use their linked official instructions for current navigation while keeping Hive's server address and OAuth sign-in flow the same.