Build a field service app in Fieldproxy from Claude or ChatGPT

A service business can describe the app it needs in plain English and have Claude, ChatGPT or another MCP client build it: the tables, the office screens, the technician's screens and the actions between them. Each save is checked against your real data, nothing is live until you publish, and a publish can be rolled back.

What field service teams ask for

  • A subcontractor compliance app: trade, insurance expiry and W-9 on file, with anyone about to lapse flagged before they are dispatched.
  • A maintenance agreement tracker: the equipment at each site, the visits each agreement promises, what is due this month and what renews next quarter.
  • A fire inspection deficiency app: deficiencies found on an inspection, the quote raised for each, and the repair visit once it is approved.
  • A parts request screen the technician fills in from the phone, with the office approving and ordering.
  • A warranty and callback log by technician and by equipment model.
  • A dispatch board for one branch or one line of business.

What you need

1. Ask for it

example request
Build a subcontractor compliance app for our commercial jobs.
- A table of subcontractors with trade, insurance expiry and W-9 on file.
- A list screen that flags anyone whose insurance expires in the next 30 days.
- A detail screen with their documents and the jobs they have worked.
- A form to add or edit a subcontractor.
Show me each screen before you publish anything.

2. What the assistant does

  • Reads the build reference first. get_widget_docs returns the app, screen and action guides and the full syntax of each widget. There are more than 110 widget types, including dispatch boards and maps, calendars, kanban boards, inspection checklists, photo annotation, signature and payment collection, parts pickers, tables, forms and charts. Web apps for the office and mobile apps for technicians have different references.
  • Creates the tables it needs. create_table and add_column are proposals: you see the table and its columns, and nothing exists until you confirm. id, created_at and updated_at are added for you. Builds only ever add tables and columns.
  • Creates the app. create_app saves it as version 1, not live. For changes to an existing app it uses save_draft (a whole app or one screen) or edit_view (a small change to one widget without resending the screen).
  • Checks it. Every save runs the linter, checks every SQL statement against this workspace's real tables without running it, and checks every navigation target. Errors refuse the save and are returned so the assistant can fix them. validate_app runs the same checks without saving.
  • Looks at it. render_view draws a screen in a real browser as you would see it, on desktop and phone, and returns screenshots, console errors, failed requests and a design audit that names each element with a problem (overflow, cut-off text, low contrast, overlapping cards).
  • Waits for you. Drafts are invisible to dispatchers and technicians. The assistant asks before publish_app, which re-checks the version and then makes it live for every user of the app.

3. If a publish goes wrong

list_app_versions shows every version and rollback_app makes an earlier one live again. A rollback is checked like a publish, so it is refused if that older version would no longer work against today's tables.

What it will not do

  • Make an app public. Apps are private to the company; customers and subcontractors get a public form, a record link or portal-user logins instead.
  • Drop a table or a column as part of a build.
  • Publish without asking.

Questions

Can Claude or ChatGPT build a field service app in Fieldproxy?

Yes. Through Fieldproxy's MCP server an administrator can have an AI assistant create the tables, build the office and technician screens and wire the actions between them. Each save is checked against the workspace's real data, and nothing is live until a person publishes it.

Is an AI-built app checked before technicians see it?

Yes. Every save runs a linter, checks each SQL statement against the workspace's tables, checks navigation, and can be drawn in a real browser on desktop and phone with a design audit. Publishing re-checks it again.

Can an AI-built app be undone?

Yes. Every version is kept, and an earlier version can be made live again with a rollback.

Does it build the technician's mobile app as well as office screens?

Yes. Web apps for the office and mobile apps for technicians are both built this way; they use different widget references, and the assistant reads the right one first.

This is one way, not the only way

This guide shows one common way to do it. A Fieldproxy workspace can be reached from every side, and anything in it can be read, changed or automated:

  • REST API: read, insert, update and delete on any table the key allows (REST API reference).
  • MCP server: 170+ tools for Claude, ChatGPT, Copilot, Cursor and your own agents, including read-only SQL, proposed writes, automations and app building (MCP tools).
  • Automations: 15 trigger types and about 50 step types, with webhooks in and signed webhooks and HTTP calls out (Webhooks and automations).
  • Public forms, record links and customer portals for people outside the company (Public forms and links).
  • Apps as JSON: new screens, fields, tables and workflows (Apps as JSON).
  • Voice and text agents that work from the same data (Voice and text agents).
  • Not listed here? Fieldproxy's team builds integrations with the other systems you use as part of setting up your workspace. Tell us what you need.

Checked against the Fieldproxy platform reference on November 17, 2026. Plain text for AI tools: /developers/guides/build-an-app-with-ai.md

Get your own custom field service management within a day. Tailored to your workflows.

We set up a workspace on your own data and connect your AI or your systems on the call.

Get Demo