Overview
The Glasscubes MCP server lets your team ask questions about your information requests directly from the AI assistant you already use, Claude or ChatGPT.
Instead of logging in and filtering the Information Request report by hand, you can ask questions like "What's overdue?" or "Which clients have bounced emails?" and get an answer in seconds.
The MCP server is read-only. It can look up and summarise your information requests, but it can't create, edit, send, chase or complete them.
What you can ask
You can ask in plain English. Your AI assistant chooses the right lookup for you.
What you want to know | Example question |
What needs your attention | "What do I need to action today?" |
Requests that have stalled | "Which requests are past due with no reminders left?" |
Delivery problems | "Which clients have bounced emails?" |
Requests by status | "Show me everything still in progress for the Smith workspace." |
Details of one request | "What's the status of the 2025/26 tax return request for Jane Doe?" |
Overall progress | "How many requests have been submitted this month?" |
Answers cover requests at any status, including Draft, On hold, Sent, Delivered, Opened, In progress, Updated, Overdue, Submitted and Verified.
Before you start: turning on MCP access
AI assistant connections are turned off by default. A Glasscubes company administrator must turn it on before anyone at your firm can connect.
Log in to Glasscubes as a company administrator.
Go to Account settings, then Advanced options, then Integrations.
Under AI assistant connections, tick Allow AI assistants to connect to this account.
If you also want external members to connect, tick Also allow external members to connect.
Let your team know they can now connect their AI assistant.
Note: Only company administrators can turn AI assistant connections on or off. Turning it off stops every connection at your firm straight away.
Connecting your AI assistant
Each person connects with their own Glasscubes login. There are two ways to add the connector.
Adding it to your own account in Claude
Open your AI assistant and go to its connectors or integrations settings.
Choose to add a custom connector.
Enter your firm's Glasscubes address followed by
/mcp/public. For example:https://yourfirm.glasscubes.com/mcp/publicWhen prompted, log in with your usual Glasscubes details.
Review the access request and click Allow.
You can now ask questions about your information requests in any new chat.
Adding it to your own account in the ChatGPT desktop app
Open the ChatGPT desktop app and go to Settings.
Select Plugins.
Click Add, then select Add MCP Server.
Enter a name for the connection, for example, Glasscubes.
Change the type from STDIO to Streamable HTTP.
Enter your firm's server address, for example
https://yourfirm.glasscubes.com/mcp/public, then click Save.Follow the prompts in your browser to log in with your usual Glasscubes details and allow access.
Adding it to your own account in ChatGPT in your browser
Log in to ChatGPT and go to Settings.
Select Plugins, then click Browse directory.
Click Add, then select Add custom MCP server.
Enter a name for the connection, for example, Glasscubes.
In the connection box, make sure Server URL is selected, then enter your firm's server address, for example
https://yourfirm.glasscubes.com/mcp/public.Read the notice, tick I understand and want to continue if you're happy to proceed, then click Create as a plugin.
Follow the prompts to log in with your usual Glasscubes details and allow access.
Once connected, start a new chat and ask a question about your information requests to check everything is working.
Adding it for your whole organisation
If your firm uses a business plan of ChatGPT or Claude, an administrator of that tool can add the connector once so it appears for everyone.
The AI tool's administrator adds your firm's MCP server address (your Glasscubes address followed by /mcp/public) in the organisation's connector settings.
Each team member opens the connector and logs in with their own Glasscubes details.
Each team member clicks Allow to complete their own connection.
Note: Adding the connector for your organisation shares the setup only. It never shares one person's access with anyone else.
Managing connections
Company administrators can see every connection at the firm under Who is connected, in Account settings > Advanced options > Integrations.
What you can do | What it shows or does |
View connections | Each person who has connected, which AI assistant they use, and when they last used it |
Revoke a connection | Click Revoke next to a person to remove their access, starting with their next question |
View the activity report | Under AI Assistant in the activity report: who asked, which lookup was used, when, and how many requests were returned |
To switch off access entirely, untick Allow AI assistants to connect to this account.
Security and permissions
Your AI assistant can only see the information requests you can already see in Glasscubes.
Your permissions apply. Every answer uses your own Glasscubes workspace access. If you can't see a request in Glasscubes, your AI assistant can't either.
No shared logins. Each person connects with their own login, so no one can see more than their own access allows.
No bulk exports. Results are limited in size, so the connector can't be used to download your whole client list.
Changes apply straight away. If someone is removed from a workspace or their connection is revoked, the change applies from their next question.
Read-only. The connector can't change anything in Glasscubes.
Good to know
The MCP server covers information requests only. It doesn't include documents, tasks, forms, approvals or the calendar.
It doesn't open or read attachments that clients have uploaded.
It's for your firm's staff by default. External members can only connect if an administrator ticks Also allow external members to connect.
Long lists come back in pages. Ask your AI assistant to "show more" to see the next set.
If a request you expect is missing, check that you have access to its workspace in Glasscubes.