Why We Let You Bring Claude, ChatGPT or Any AI to Noloco

Your AI assistant can now build your Noloco apps, and your team can use AI to work with them.

It's Thursday afternoon and the operations lead at a facility services company has three questions waiting: a sales rep wants to know if the Martinez permit came back, a subcontractor wants to know which jobs start Monday, a homeowner has emailed to ask when the panels go on the roof. All three answers are already in the system, but she's the only one who knows where to look, so she stops what she's doing and looks.

I've watched versions of that afternoon in consultancies, accounting firms and construction teams for years. The business has the information it needs, the problem is that getting to it still depends on one person.

This month, we shipped two changes that start to fix that. I think they're the most important things we've released since Nola, our AI app builder, and I want to explain why.

TL;DR

• From today, you can use your existing Claude, ChatGPT, Codex, Copilot, Cursor, Windsurf or Gemini subscription to build Noloco apps through the Noloco MCP.
• You're no longer limited by how many AI credits you can afford. Connecting your own AI assistant is included on every plan.
• We've also introduced the App MCP. It lets everyone on your team ask their own AI assistant questions about your app's data. Your clients can do the same, if you allow it. Each person only sees what their role permits.
• Admins decide which roles can connect and whether they get read or read-and-write access. AI assistants can never delete records.
• The idea is simple: one system for your team, your clients, and the AI agents helping run your business.

What did we actually ship?

Two MCP servers, and they do two very different jobs.

MCP (Model Context Protocol) is the standard way AI assistants connect to other software. Instead of copying information into a chat window, your assistant can work directly with the real data in your tools.

Here's the simplest way to think about the two:

• Noloco MCP: Your AI assistant helps you build and change your app.
• App MCP: Your AI assistant helps you use your app, with the same permissions you already have.

The Noloco MCP: Build your app with the AI you already use

The Noloco MCP is for the people building your apps. Connect your AI assistant and ask it to turn your data into a client portal, add a form for new requests, or send a Slack notification when a deal is won.

It can create tables, pages, views, forms, workflows, roles and permission rules. It gets access to the same building tools as Nola, our own AI builder.

The App MCP: Let everyone use their own AI assistant

The App MCP is for the people using your apps.

Each person connects their own AI assistant to a Noloco app. They can ask questions, find records, and add or update records if their role allows it. They can't change how the app is built, everyone gets the access they needs, and nothing more.

Why are we letting you build with the AI you already pay for?

When we relaunched Nola in August, we introduced AI credits to our plans. The more you used Nola, the more credits you needed. That was a practical decision: AI is expensive to run, and someone has to pay for it.

But we never planned to build the business around credits. When a company buys Noloco, it's paying for peace of mind. It wants a system that's secure and reliable, with real people behind it. It's not buying tokens. And we never wanted to become a middleman for Anthropic or OpenAI.

So now, if you already pay for Claude, ChatGPT, Codex, Copilot or Cursor, you can connect it to Noloco and start building. Connecting is included on every plan.

The only building tools that still use Noloco AI credits are the ones running on our own AI, such as generating Canvas pages, record page sections and workflow actions.

Credits still have their place, they just aren't the gate anymore. This makes Noloco much more accessible: How much you can build is no longer tied to how many AI credits you can afford.

There's another reason we wanted to do this. Most AI app builders have a business model built around the AI customers use to build their apps. Letting customers bring their own AI would mean giving up a big part of what they're selling.

We made a different choice. We'd rather be judged on whether your business runs well on Noloco than on how many tokens pass through us.

What does building with your own AI assistant actually look like?

Say you run operations at a 40-person consultancy. Your project information lives across a handful of tables, and clients keep asking for updates.

You open Claude and type: "Turn the Projects table into a client portal where each client sees only their own projects."

Your assistant reads your tables, builds the pages and sets up the permission rule.

You stay in control throughout:

• ‍Changes don't go live without you. Pages, views, forms, buttons and workflow actions stay as drafts until you publish them. The assistant checks with you before publishing or deleting anything.‍
• Access has to be approved. Admins decide which apps an assistant can access. Building also requires read-and-write access, plus builder and data admin permissions.‍
• No API keys sitting in config files. You connect with a one-time sign-in. Access is checked on every request, so removing someone's builder role takes effect immediately.

There's one important thing to understand: a builder's assistant acts with the data admin role. That means it can read and change any record in the apps you've approved.

That's the right level of access for someone building the app. It's the wrong level for everyone else. And that's where the second update comes in. Honestly, it's the one I'm most excited about.

Why has giving your team AI access to company data been all or nothing?

Most growing businesses have their operational data sitting in a database, an Airtable base or a badly formatted spreadsheet.

They'd love to give everyone in the company access to that information through AI. But there's a problem: giving an AI assistant access to company data often means giving it access to everything. Either one connection can see all your records, or nobody gets access at all.

People aren't waiting for a better option. A 2026 PagerDuty survey found that at companies with fewer than 1,500 employees, 40% of workers had entered customer data into public AI tools. The same survey found that 66% of workers who had used AI at work did so even though they believed company policy didn't allow it.

That's what happens when the safe answer is simply "no". People find another way, usually by exporting data and pasting it into a chat window.

I don't blame them, the tools haven't given them a sensible middle ground.

How does the App MCP keep everyone within their own permissions?

The App MCP gives everyone the same access in their AI assistant that they already have inside their Noloco app.

The assistant signs in as the person who connected it. It can access exactly what their role allows, down to individual records and fields. Nothing more.

Admins decide which roles can connect in the first place, so you stay in control.

Each role has an Agent access setting:

• None: The role can't connect an AI assistant.
• Read: The assistant can find records and answer questions.
• Read & write: The assistant can also create and update records.

Roles that can see all data start with Read & write. Every other role starts with None. That means nothing is automatically opened up to clients or contractors. You decide when to switch it on.

There are also a few limits that apply regardless of the role's permissions:

• AI assistants can't delete records, even if the person connecting them has permission to do so.
• Each connection covers one app only. It doesn't include any tools for changing how the app is built.
• Activity counts as that person's activity and uses their seat.
• Reads and writes go through the Noloco API, so your plan's API rate limits still apply.

Let's go back to our facility services company from the beginning:

- The sales rep asks Claude whether the Martinez permit came back. Claude checks the jobs he's assigned to and gives him the answer.

- The subcontractor asks ChatGPT which of her jobs start Monday. She sees only her own jobs.

- The operations lead turns on Read access for the Customer role. Now the homeowner can ask when his panels are going on the roof and get his installation date. He sees nobody else's information.

- And the operations lead doesn't have to stop what she's doing to answer any of those questions.

What does this change for the person keeping the operation running?

In most growing businesses, there's one person who acts as the switchboard. They built the tracker, they know which spreadsheet holds what, they know where the latest update lives and who needs to be asked. Every "where are we on this?" lands with them. The business works because they keep answering.

The App MCP doesn't replace that person. It takes the repetitive lookup questions off their plate. Colleagues and clients can find answers themselves, using the AI assistant they already have, within the same permissions the app already enforces. The operations lead sets the rules once and stops being the only way into the system.

That's also why we built two separate servers. The person designing the operation needs the tools to build and change the app; while everyone else needs access to the information relevant to their own work. Those are two different jobs. Mixing them is how companies end up back at all-or-nothing access.

If you're thinking about how to set up access properly, we've put together a guide to [client portal permissions]. The same roles now determine what each person's AI assistant can see.

What if your data lives in Airtable or a spreadsheet?

You don't need to move it first. Noloco connects to Airtable, Google Sheets, Postgres and other data sources, so your data can stay where it is.

You define roles in Noloco, and those permissions carry over to every connected AI assistant.

Maybe you're happy with Airtable as your data layer. In that case, this is a straightforward way to give more people access to your company data through AI, without rebuilding everything.

Or maybe you're starting to hit its limits: clients need their own logins, permissions are getting harder to manage, and you're paying more every time you add another person who needs access. If that's where you are, I walk you through what to do when Airtable is holding you back in the latest article How to migrate from Airtable, you don't have to move all at once.

Final thoughts

AI assistants are already becoming part of how people work. The question is whether they'll work with your company's data safely, or find ways to work around the systems you've put in place.

We built these two updates so you don't have to choose between giving people access and keeping control: Build with the AI you already pay for, let your team and clients use AI to find and update the information they need, and keep everyone within the permissions you've already set. And stop making one person responsible for finding every answer.

That's what we're working towards with Noloco: one system for your team, your clients and the AI agents helping run your business.

Ready to try it? Connect your AI assistant and start building, or show your team how to connect their own assistant to a Noloco app.

If you'd rather see it working with a setup like yours first, book a demo with me.

Frequently asked questions

What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to other software. An MCP server is the connection point a tool provides, allowing an assistant like Claude or ChatGPT to read and update information directly, rather than relying on you to copy and paste data into a chat.

What's the difference between an MCP for building an app and an MCP for using one?

A building MCP lets an AI assistant change the app itself: its tables, pages, forms, workflows and permissions. A usage MCP lets the assistant work with the app's records as a specific person, within that person's permissions. Most businesses need the first for the people building and managing their apps, and the second for everyone else.

Is it safe to connect an AI assistant to company data?

It depends on how access is controlled. A safer setup has the assistant sign in as a specific person and limits it to what that person's role allows. Access should be checked on every request, and admins should be able to revoke it.

The risky alternative is one shared connection with access to everything. That's exactly the problem the App MCP is designed to avoid.

What's the difference between read and read-and-write access for an AI assistant?

With Read access, the assistant can find records and answer questions, but it can't change anything. With Read & write access, it can also create and update records.

A sensible starting point is to give most roles Read access and reserve Read & write for people who already have permission to edit that data.

Can clients use their own AI assistant to check on their work?

Yes. An admin needs to enable Agent access for the client role. Once connected, the client's assistant can only see the information that client is allowed to see.

For example, a client could ask about their projects, invoices or installation dates without getting access to anyone else's information.

Do I need to pay for AI credits if I build with my own AI assistant?

Not for the work your own AI assistant does. Your existing Claude, ChatGPT or other AI subscription covers that.

In Noloco, only the building tools that run on our own AI use credits. That includes generating Canvas pages, record page sections and workflow actions. See our pricing page for the details.