AI integration · Custom MCP servers

What if you could manage your business by having a conversation?

Update website content in ChatGPT. Ask your CRM a question in plain English. Connect your own apps to AI without handing over the keys. I build secure, practical integrations around the software you already use.

Your software should be easier to use.

You have a website, a CRM, perhaps a booking system, a database and several spreadsheets. Each one comes with its own menus, login and way of doing things. Even a simple task can turn into five screens and a long set of instructions.

Imagine opening ChatGPT or Claude and saying, “Show me the artworks still available”, “Find customers we haven’t followed up”, or “Update this page, but let me approve the change first.”

An AI-powered business system makes that possible by connecting a conversational interface to carefully selected functions in your existing software. You describe the outcome. Your authorised backend performs the work. You stay in control.

I build these connections at Erzo, combining custom web development, APIs and responsible AI. The goal is not another complicated platform. It is a more human way to use the systems you already rely on.

What could you manage through AI?

Start with a task your team already does. Make that one thing easier, safer and quicker to complete.

  • Website content management

    “Change this service description, then show me a preview.”

    Give approved editors a simpler way to update pages, artwork, product descriptions, images and availability. Changes can be saved as drafts and reviewed before publication. The site’s layout and code can remain developer-controlled.

  • CRM and sales

    “Which leads have not been contacted this week?”

    Search customer records, summarise pipeline activity, prepare follow-ups and update permitted CRM fields. The integration works through the CRM’s supported interfaces and your organisation’s access rules.

  • Custom applications

    “Show the outstanding requests by department.”

    Add conversational access to a customer portal, internal app, booking system or operational database. Staff can ask for information without navigating several screens, while the application remains the source of truth.

  • Reporting and insights

    “Compare this month’s enquiries with last month.”

    Retrieve information from approved systems, apply reliable calculations and present meaningful summaries. Decisions are grounded in real records rather than invented figures.

  • Workflow automation

    “Prepare follow-up tasks for overdue enquiries.”

    Connect approved actions to existing workflows, from drafting correspondence to starting onboarding processes. Sensitive steps can remain subject to human approval.

Examples describe potential integrations, not features automatically available in every software product.

How does it work? Meet MCP.

MCP stands for Model Context Protocol. It is a standard that allows compatible AI applications to discover and use tools supplied by other systems. Think of it as a controlled bridge between the conversation and the software where your information lives.

For example, suppose you run an online catalogue. You say, “Mark White Lilies as sold.” The AI selects the appropriate catalogue tool. Your MCP server checks the user’s access, identifies the artwork, validates the change and saves a draft. You can review it before asking the system to publish.

The AI is the interface. Your application remains the authority. It decides what is allowed, what gets saved and what goes live. That distinction makes this more useful than a chatbot that merely writes suggestions, and much safer than handing a model unrestricted access to your database.

MCP is not the only integration method. Where a direct API, automation service or purpose-built conversational interface is more suitable, I will recommend that instead. The objective is a reliable business outcome, not a particular acronym.

Why a conversation instead of another dashboard?

Dashboards are excellent for reviewing lots of records or performing highly structured work. But they are not always the quickest way to answer a question or make one small edit. A conversation can remove unnecessary navigation and training.

  • Less learning. Describe the job instead of remembering which menu holds the right form.

  • Useful access to information. Ask for approved records, summaries and reports in everyday language.

  • Existing software stays in place. Your website and CRM do not necessarily need replacing.

  • One consistent interface. Compatible AI clients can use a shared set of carefully defined business functions.

  • Human approval. Changes can be drafted and previewed before anyone publishes or sends them.

I keep conventional interfaces where they make sense. A conversational CMS should complement your application, not force every task through a chat window.

AI access without unrestricted access.

Connecting an AI assistant to company systems deserves the same care as any other production software. I design integrations around least-privilege permissions, secure authentication, validated inputs and clear approval steps.

An editor can be allowed to change website descriptions without touching its layout or code. A salesperson can search approved CRM records without permission to export everything. A manager can approve publication while other team members prepare drafts.

Where appropriate, I build revision history, audit logs, error handling and rollback into the workflow. I also review what information must be sent to an AI provider and what should stay inside your infrastructure. For South African businesses, personal-information handling may require a POPIA assessment.

Security rules belong in the backend, not in a prompt. An instruction like “please don’t delete records” is not a substitute for a server that refuses deletion.

How I build your integration

A focused first use case, clear permissions and realistic testing come before adding more AI features.

  1. Start with one valuable task

    We talk about your current software, where your team gets stuck and which repeatable request would genuinely save time. I recommend starting small rather than connecting everything at once.

  2. Assess the systems

    I check the available APIs, authentication, data models, hosting and privacy requirements. Some existing tools are ready to connect; others need a small integration layer or an alternative approach.

  3. Define permissions and approvals

    We agree what the assistant may read, what it may change, who can use each action and when a person must approve a result. These boundaries live in server-side code.

  4. Build and test the integration

    I implement the MCP server or another suitable API-based connection, then test actual requests, wrong inputs, failed operations and access controls with the AI client you intend to use.

  5. Launch and evolve

    Once the workflow is reliable, I help configure access and explain how to use it. More capabilities can be added as the business sees value.

Will it work with your website, CRM or app?

Modern applications built with Next.js and other frameworks are good candidates when their business functions can be exposed through safe APIs. Many CRM platforms and established content management systems also have interfaces that can be connected without rebuilding the product.

Older software may require an adapter, a restricted integration service or a different approach. The answer depends on the system you have, the data it holds and which actions you want available.

For a content-heavy website, I can build an AI-powered CMS that allows approved editing of titles, descriptions, images or availability while keeping site design and technical configuration under developer control. For a business application, the same principle can expose searches, reports and carefully scoped workflows.

ChatGPT, Claude and other AI clients may have different plan requirements, authentication flows, image capabilities and write-tool restrictions. I test compatibility for your intended setup rather than assuming every feature works identically everywhere.

If you are starting a new site, see my website development service. If your priority is automatic processing rather than a conversational interface, read about workflow automation.

Who is this for?

Business owners who want to update their own sites without learning an admin system. Marketing teams that need controlled publishing. Sales teams that spend too long searching for customer information. Organisations with custom applications that could be easier for staff to operate.

The best fit is usually a business with valuable existing software and one clear, recurring task that takes more effort than it should. You do not need a huge AI transformation programme. You need a practical place to begin.

Common questions

Do I need to rebuild my website or replace my CRM?

Usually not. If your system has suitable APIs or a safe way to expose business operations, I can often add a conversational interface around what you already use. I assess the existing software before recommending a build.

Can I use my existing ChatGPT or Claude account?

Possibly. MCP support, connection options, write permissions and authentication can vary by AI platform and account plan. I verify the required features for your specific client before promising a workflow.

Will AI be able to change anything in my business?

No. The server exposes only approved operations, with user permissions and validation enforced in the application. We can require a draft, preview and explicit confirmation before changes go live.

Can different team members have different access?

Yes, where the underlying software supports the necessary identity and permission model. One person may only read information, another may prepare drafts and an authorised manager may publish.

What is an MCP server?

MCP stands for Model Context Protocol. It is a standard that lets compatible AI applications discover and use tools offered by external systems. The server connects requests from an AI interface to specific, controlled business functions.

Can this work with WordPress or an older application?

Sometimes. WordPress and many other products provide APIs, while older systems may need a custom adapter. I recommend the safest workable integration after reviewing what you have.

What does an AI business integration cost?

Pricing depends on the systems involved, the number of actions, access controls and ongoing support. An Ideas session is a practical starting point: we define a first use case and I can propose a clear scope and cost.

What if your software simply understood the task?

Tell me which website, CRM or application you use, and one thing you wish were easier. I’ll help you decide whether an AI integration is worthwhile. An Ideas session is a good place to turn it into a concrete plan.

Talk about your AI integrationor email henry@erzo.co.za

South African time (SAST, UTC+2). I reply to every message myself, within 24 hours.