Product 7 min read

Build it where the conversation happens: FormWork meets MCP

Connect FormWork to Claude, ChatGPT or another MCP assistant. Turn the context in your conversation into forms, approvals and automations you can review and run.

Andrius Bartulis

By the time someone opens a form builder, a lot of the useful thinking has already happened somewhere else.

An agency has discussed the client’s onboarding process. A team has worked through the exceptions to its purchasing policy. Someone has compared three versions of a brief and finally understood what the form actually needs to do. Increasingly, that work happens in a conversation with Claude or ChatGPT.

Then comes the translation. Open another tool. Add the questions. Recreate the rules. Remember which requests need approval, who should receive the email, and what happens when the customer comes back with a correction.

Today we’re introducing FormWork’s MCP server. You can connect FormWork to an AI assistant and build the solution from that conversation: forms, shared data, stages, approvals, automations and the messages around them. You can also ask about the entries you’ve received and work through the requests waiting for you.

This is a substantial step towards making FormWork AI-first. The context you’ve built up with your assistant can become something your team actually uses.

Build from the brief

Say you’re building a supplier onboarding process for a client. Your conversation already contains the brief: what Purchasing needs, which documents Finance checks, and which suppliers require an extra review.

You might ask:

In Acme’s Supplier onboarding project, build the intake form from this brief. Suppliers should be able to add several contacts. Give Purchasing the first review, send higher-risk suppliers to Compliance, and email the supplier when we’re ready to proceed. Prepare a draft and show me what needs deciding.

The interesting part is everything around the questions. A repeating list of contacts is useful, but so is knowing who owns the request after it arrives. So is keeping internal review notes away from the supplier. So is being able to change the process later without making old submissions impossible to understand.

Those are the things FormWork already does. The MCP connection gives the assistant a way to assemble them together, and to read what’s already there before changing it. For common processes, it can start from a blueprint and adapt it, asking you to choose the real people and groups in your account.

You still have decisions to make. “Higher risk” needs a definition. Compliance needs an actual approver. An assistant that asks those questions is more useful than one that quietly invents the answers.

Why build on FormWork?

AI makes it much easier to write a form from scratch. We’ve written about that choice, and there are good reasons to take either route.

But generating the first version is only part of the job. Someone has to run it. Answers need validation. Files need access controls. Emails need to be delivered. A request waiting for approval needs an owner and a place to wait. When the policy changes next month, the team needs to know what changed and what happens to work already in progress.

With this connection, the assistant builds on FormWork’s existing form engine and process model. Calculations and validation run on the server. Published versions preserve the questions an entry was answered with. Stages hold the work people are doing, and automation runs show what happened when something went wrong.

That foundation matters just as much when AI does the building. Perhaps more: a plausible-looking result is easy to produce, and easy to mistake for a finished solution.

A draft you can inspect

We’ve kept the distinction between preparing a change and making it live.

Questions, team fields, stages and automations are built in drafts. The assistant can test an automation with sample answers and show what would happen, without sending an email or changing a real entry. In clients that support MCP Apps, you can preview the form and see the publish review inside the conversation.

The review includes the consequences for existing entries and the places data would be sent. Adding a confirmation email to the right person is helpful. Adding one to the wrong address is a problem, even if every step is technically valid.

You review, then publish. If a teammate has changed the draft since that review, FormWork refuses the stale publication so it can be checked again. You can also open the draft in the regular builder and work on it there.

Templates, views and data table columns have their own rules and can change immediately; the docs explain those boundaries. They matter when choosing how much access to give an assistant.

Built for agencies with more than one client

For an agency, the conversation often spans several projects. FormWork accounts keep clients separate, and one assistant connection can cover the accounts you choose.

Every change names its account and project. If it’s unclear which client you’re talking about, the assistant is instructed to ask. FormWork checks the resources against that account and the connection’s allowed projects too.

You can reuse a solution in another client’s project without copying the first client’s form entries. The new client’s people, approvers and secrets still need choosing. That makes a working process a useful starting point for the next project, while keeping the client-specific decisions explicit.

Each client’s admins remain in control. They can allow assistant work, prevent publishing, make access read only, or turn it off. They can remove their account from a connection without affecting your work for anyone else.

The conversation can continue after launch

Once the form is live, you can ask:

Which supplier requests are still waiting for Compliance, and how long have they been there?

Or:

This supplier didn’t get its confirmation email. Show me what happened.

The assistant can search entries, summarise them, read their history and inspect automation runs. It can help prepare a fix, test it, and bring it back for review.

It can also help with live work: assigning requests, moving them through stages and deciding approvals sent to you. Bulk entry changes start with a dry run showing the targets and effects. That is particularly useful for a request such as “close the old tickets”: you should see which tickets “old” actually selected before anything moves.

Your assistant, your access

MCP is the Model Context Protocol: a way for an assistant to use tools provided by another application. Here, FormWork provides the tools and your chosen assistant uses them.

You sign in to FormWork and choose the accounts, projects and permissions the connection can use. It acts as you, within those limits and the account’s settings. An assistant connected for reporting doesn’t need permission to publish or change entries.

The data it reads goes to your assistant’s provider under your agreement with that provider. That includes entry content if you’ve granted access to it. FormWork withholds passwords, secrets and private file and approval links from MCP results, but choosing the right access still matters.

Where we’re heading

We want FormWork to fit naturally into the way teams are starting to work. If you’ve spent an hour thinking through a client’s process with an assistant, you should be able to carry that understanding into the build, then keep using it when the process needs attention.

MCP gives us a practical way to do that today. The conversation becomes another way to work with FormWork, alongside the builder and the API, using the same drafts, validation and permissions.

To try it, open AI assistants in FormWork’s sidebar and copy the server address into your assistant’s custom MCP connection. The connection guide walks through setup, and the overview covers what you can build and do.

I’d like to hear where it helps, and where you still have to translate too much yourself. That will tell us what to build next.

— Andrius

AIMCPagenciesautomationsproduct

Keep reading