Building Extensions

Connect your own AI agent. It builds the extension for you.

The idea

Sois AI is extensible. Every workspace ships with built-in modules: contacts, tasks, inbox, documents, files, departments. But what if you need a CRM pipeline, an invoice tracker, a project board, or something entirely unique to your business?

You build it with the AI agent you already use. Connect your agent (Claude, ChatGPT, Cursor, Gemini, or any MCP-capable tool) to the SOIS MCP build tools, describe the app you want, and your agent builds, validates, and ships it. There is no in-app builder wizard to learn, and you do not need the developer portal to build a custom extension. Connecting your agent is enough.

Because your own AI does the reasoning and the work, SOIS charges you nothing to build.

How it works

1. Connect your agent

Connect your agent to your workspace's MCP server (OAuth is the recommended path, with nothing to paste; a token also works). The build tools are part of that same server and appear automatically for builder-capable accounts. See Build with your Agent for the connection details and the full tool list.

2. Describe what you want

Tell your agent what the extension should do, what data it stores, who can access it, and whether it needs credentials or external APIs. Your agent reads the canonical extension rules over MCP and plans the build.

3. Your agent builds and validates

Your agent generates everything: the data model, the UI, the agent toolkit, permissions, and translations. It runs the free static validation gate as often as it likes and fixes issues locally until the bundle is clean.

4. Upload, review, publish

Your agent uploads the bundle. It is hardened, gated, and queued for review. Once approved, install it from the extension store inside any Sois AI workspace. It appears in the sidebar, the explorer, the search bar, and the agent's toolkit, instantly.

What gets built

Each extension is a self-contained bundle:

Component What it does
Data model Tables, columns, and JSON fields scoped to your extension
UI A fully interactive interface that renders inside the workspace
Agent tools Functions the AI agent can call to read, create, update, and search your extension's data
Permissions Granular access control: who can view, edit, delete
Translations Multi-language support out of the box
Dashboard widget Optional at-a-glance metrics on the workspace dashboard

Example extensions

These are the kinds of apps your agent can build from a plain-English brief.

CRM Pipeline

"I need a CRM with leads, stages (New, Qualified, Proposal, Won, Lost), and the ability to drag leads between stages. Each lead has a company, contact, value, and expected close date."

Your agent builds:

  • A leads table with stage tracking
  • A Kanban-style drag-and-drop UI
  • Agent tools: searchLeads, createLead, updateLeadStage, getLeadsByStage
  • Dashboard widget showing pipeline value by stage

Invoice Tracker

"An invoicing extension. Each invoice has a client, line items, tax, total, status (draft, sent, paid, overdue), and a due date. I need to generate PDF invoices and email them."

Your agent builds:

  • An invoices table with line items stored as JSON
  • A detail view with line item editor and status controls
  • Agent tools: createInvoice, sendInvoice, markAsPaid, listOverdueInvoices
  • Dashboard widget showing outstanding vs paid totals

Equipment Booking

"A booking system for shared equipment. Staff can reserve items for a date range. No double-booking. Show a calendar view."

Your agent builds:

  • An equipment table and a bookings table
  • A calendar UI with conflict detection
  • Agent tools: checkAvailability, createBooking, cancelBooking, listBookingsForDate

Client Portal Content

"A knowledge base for our clients. Articles organised by category. Clients can search and browse. Admins can create and edit."

Your agent builds:

  • An articles table with categories and rich text content
  • A searchable, categorised browse view
  • Agent tools: searchArticles, createArticle, updateArticle
  • Permission split: clients get read-only, admins get full CRUD

Who can build

Anyone can connect an agent and build. You do not need a developer account or the developer portal to create a custom extension for your own workspace.

Builder What they do
A tenant user Connect your agent, build an extension for your own workspace, and route it privately to yourself with a deployment code.
An external developer Connect your agent, build for the ecosystem, and publish to the marketplace to earn from every use.

Both paths use the same MCP build tools and the same bundle format. The only difference is where the finished app goes.

Deployment options

Option How it works
Public store Available to all Sois AI workspaces globally. Earn revenue when others install and use your extension.
Private deploy Generate a deployment code in your workspace and route the extension directly to you. No one else can see it.

Agent integration

Every extension automatically gets agent tools. When the extension is installed, the workspace AI agent (and any external agent connected over the MCP server) can interact with its data, with no extra setup.

Ask the agent:

  • "How many open leads do we have?"
  • "Create an invoice for Acme Corp, 10 hours of consulting at $150/hr"
  • "Book the projector for Friday 2-4pm"
  • "Find all articles about onboarding"

The agent uses the extension's tools just like it uses the built-in modules. Search, create, update, summarise, all through natural language.

Get started

  1. Read Build with your Agent for the MCP connection and the build tools.
  2. Connect the agent you already use.
  3. Describe your idea, let your agent build and validate it, then upload it.
  4. Deploy it privately to your own workspace, or publish it to the marketplace and earn from every use.

No in-app wizard. No developer portal required. Just connect your agent and describe what you need.