How to connect ChatGPT to your CRM and use it as an AI analyst: The Uspacy example

Which deals require attention? Which clients haven't been contacted in a long time? Which tasks are overdue? After connecting ChatGPT to Uspacy through MCP Server, you can ask these questions in plain English. Uspacy remains your central Workspace where your business data is stored and managed, while AI helps you analyze that information faster.

Without an integration, you have to export data, upload it into ChatGPT, and explain the structure of your tables each time. MCP streamlines this process by allowing ChatGPT to access the available tools in Uspacy and generate responses based on your current CRM data.

— Review my open deals in Uspacy and identify any that don't have a next activity scheduled. Don't make any changes.

You can then refine your request naturally—for example, limit the results to high-value deals, change the date range, or group the findings by account manager. Instead of running one-off searches, you can have an ongoing conversation that builds on previous results and supports deeper analysis.

MCP Server is a secure bridge between AI and an external system. It enables ChatGPT to use authorized tools to retrieve data or perform supported actions within Uspacy.

The assistant operates within the permissions of the authenticated Uspacy user. It can access only the records that user is authorized to view, while editing capabilities depend on the user's permissions and the available MCP tools.

The integration supports CRM, tasks, activities, comments, search, and user management. It's best to start with read-only requests so your team can get comfortable reviewing AI-generated responses before enabling any actions that modify data.

To connect ChatGPT to Uspacy, use the Codex desktop application. In Codex, you'll need to add the Uspacy MCP Server and authenticate with your Uspacy Workspace.

For your first test, use this safe prompt:
— Check whether you have access to the Uspacy tools. Don't create or modify anything. Briefly describe what data you can access.

Then try a simple practical task:
— Show me my next five upcoming tasks in Uspacy. Include the task name, deadline, and status. Don't edit anything.

If Codex returns current data from Uspacy, the connection is working correctly. From there, you can start analyzing deals, tasks, and your sales funnel.

A prompt like "Analyze our sales" is too broad. High-quality responses require a clearly defined role, data source, time period, evaluation criteria, output format, and explicit instructions about what actions the AI is allowed to take.

Universal prompt template:
— Act as a sales analyst. Review [data type] in Uspacy for [time period]. Identify [a problem or opportunity]. Support each finding with evidence from the CRM. Present the results as a [table, list, or summary]. Don't make any changes.

Practical use cases:

• Team meeting preparation. Identify the five situations that require the team's attention today.
• At-risk deals. Find deals with no activity for more than seven days or no next scheduled activity.
• CRM data quality. Identify deals that are missing an owner, deal value, close date, or next step.
• Repeat sales opportunities. Find clients with closed deals who haven't been contacted in a long time.
• Manager performance review. Group high-priority issues by deal owner and explain why they should be prioritized.

After the initial response, refine your request to narrow the analysis—for example: "Show only deals over USD 50,000," "Group the results by account manager," or "Prepare a 20-minute meeting agenda." These follow-up prompts make the conversation far more valuable than a static report.

ChatGPT can only see what is recorded in Uspacy and available to the user. If a manager didn't log a call, update a deal stage, or schedule a task, that context will not be included in the analysis.

Ask the AI to separate facts from interpretations:
— For each finding, show the facts from Uspacy, your interpretation, and any missing information needed to make a confident decision.

Safe workflow:

• Start by reviewing the data.
• Request a list of recommended changes.
• Review each suggestion.
• Approve edits separately.
• Use one-time permissions.
• Verify the results in Uspacy.

A useful follow-up step is: "Review your previous analysis. Find any data that could contradict your conclusions." The final decision always remains with the user, while AI serves as an analytical assistant.