Many “build a CRM with AI” tutorials end with a prototype: a page that looks like a CRM, with data that lives in the browser or in a spreadsheet. Here the agent works on a real business application platform. Claude connects to a NET Genium Online instance through its built-in MCP server and builds the CRM straight into it – the tables, the forms, the relations between them and the overviews – in one conversation. The result runs in production the moment it is written, and you keep developing it with the same agent.
What the agent built in one conversation
The video starts with an empty instance and a request written the way a customer would say it, not as a specification. Nothing between the request and the result was cut out; the long passages are only sped up, with the real elapsed time on screen.
- A connected application, not three loose tables. Companies, contacts and a log of activities, each with its own form and overview. On a contact, the company is a real foreign key chosen from a list, and the company form has tabs with its contacts and its activities.
- Searching and filtering come with it. The overviews can be searched and filtered without anyone setting them up.
- A change to a running application. Asked for annual revenue and a flag for key clients, the agent added both fields to the existing application and placed them on the form and in the overview. Nobody rebuilt the application or migrated the database by hand.
- Data written by the agent itself. Eight companies, sixteen contacts and twenty-four activities from the last two months – after the agent first listed its plan and waited for confirmation.
Building the application took seventeen minutes, filling it with data six minutes.
Why a real database matters to an agent
An agent is only as useful as the system it is allowed to touch. In NET Genium it does not get a thin API over a row-limited base, but the instance itself:
- A real SQL data model – tables, columns, data types and required fields, so the agent writes against real names instead of guessing them.
- No limit on records – the only ceiling is the storage space of your plan.
- The application model, not just the records – the agent creates and changes forms, view pages, controls, queries and scripts, so the CRM keeps growing with the business instead of staying a first draft.
- Permissions you grant explicitly – reading and writing are two separate permissions on the consent screen, and the access can be revoked in the instance at any time.
Build it yourself
- Create an instance. With one user it is free with no time limit, with no credit card and nothing to install.
- Connect Claude. In Claude, add a custom connector with the address https://mcp.netgenium.online/api.aspx, enter the name of your instance, sign in to it and allow reading and writing. The whole setup is described in the guide Connecting an AI tool to the MCP server.
- Describe your CRM. Write how your company works, not how the software should look – the prompt below is a good start.
- Review the proposal and let it write. The agent shows what it is going to create; once you agree, the application appears in your instance.
- Keep going. Ask for new fields, overviews or sample data the same way – the agent changes the running application instead of starting over.
Build a CRM in my NET Genium instance. We track companies (name, city, country, industry, number of employees, customer since, notes), the contacts at each company (first name, last name, e-mail, phone, role) and a log of activities with a company and optionally a contact (type call / e-mail / meeting / demo, date, subject, description, done). Show the contacts and the activities on the company form. Propose the structure first, then write it.
More ready-made prompts for six common business areas are in the Prompt gallery, and the way an agent works with an instance step by step is described in Agentic development.
What it costs
One user is free for good. Each additional user costs $12 (€10) a month, with no limit on records and the data stored in the EU. The full comparison is on the page Pricing.