Bring your own AI. CanopyBay keeps it safe.
MCP (Model Context Protocol) is the open standard that lets AI assistants use other apps. CanopyBay gives your AI a small set of safe tools. Your AI does the asking and the preparing. People do the approving.
- ClaudeAisha, HR
- ChatGPTNur, owner
- GeminiRaj, finance
- Local AIIn your building
- CanopyBay chatEveryone else
MCP
CanopyBayTools your AI can use
search_recordsprepare_draftcreate_workflowpreview_change
Approval Only in CanopyBay, by a person
Your AI can
- Find and read what you’re allowed to see
- Prepare replies, invoices, rosters and reports
- Change forms, fields, workflows and pages, with a preview first
- Build a new sub-module from ready-made blocks
Your AI can’t
- Approve anything. Approvals happen in CanopyBay
- Move money or change bank details
- See other people’s records
- Touch CanopyBay’s code, or obey instructions hidden in a record
How it works
Five steps, from switching it on to a change going live.
Aisha asks ClaudeAdd a Head of Ops approval step to leave requests over three days.
Claude, using CanopyBayI’ve prepared the change to your Leave workflow. Preview it in CanopyBay and approve it there. Nothing changes until you do.
Turn it on
Your admin switches on MCP and picks which AI assistants staff may use.
Connect your AI
Each person adds CanopyBay to Claude, ChatGPT or Gemini and signs in with their own CanopyBay login.
Ask in plain words
“Add a Head of Ops approval step to leave over three days.” “Reply to last night’s enquiries.”
Your AI prepares it
It uses CanopyBay’s tools to draft the work, or a change with a preview link. Nothing is live yet.
A person approves it in CanopyBay
Never in the chat. Changes go to a test copy first, then live, and can be undone.
The tools your AI gets
A small set of high-level tools. The rules live in CanopyBay, not in the AI, so even a vague request can only produce a safe, valid result.
| Tool | What it does | Needs approval |
|---|---|---|
search_records | Finds records the person is allowed to see | No |
describe_schema | Lists the forms, fields and modules that exist | No |
prepare_draft | Drafts a reply, invoice, roster or report for review | Yes, before it’s sent |
add_field | Adds a field to a form or record | Yes |
create_workflow | Builds a new approval or automation flow | Yes |
create_page | Adds a page or a view for your team | Yes |
preview_change | Returns a link to preview any change in CanopyBay | No |
check_status | Checks a long job, like a big import | No |
Built so a mistake can’t go live
- Each AI has its person’s permissions.
- A part-timer’s AI can’t see payroll.
- Approvals only in CanopyBay.
- A “yes” in a chat is never enough.
- Records are read as information.
- Text inside a record, like a customer’s message, can’t give your AI orders.
- No destructive tools.
- Nothing can be deleted through MCP, and every change can be undone.
- Configuration, not code.
- Your AI changes how CanopyBay works for you, never CanopyBay’s code.
- Everything is logged.
- Who asked, which AI, which tool and who approved it.
- You stay in control.
- Admins choose the allowed AIs and can switch off anyone’s access at once.
Local AI. No token bills. No AI subscriptions.
Most AI work in a business is everyday work: drafting replies, reading receipts, matching payments, writing reports. It runs well on one AI computer in your building. We install it and tune it, your whole team shares it, and you stop paying per question or per person.
- Cloud AI, pay per use
- Every question is billed in tokens. The busier the month, the bigger the bill.
- AI subscriptions
- A monthly fee for every person who needs AI, whether they use it a lot or not.
- Local AI
- One machine in your building. No bill per question and no fee per person. Worth it once your team uses AI every day, and we size it with you before you buy.
Language
- Drafts and replies for enquiries, parents and customers
- Reads and summarises contracts, policies and long threads
- Voice to text for site notes and meeting minutes
- Answers questions from your own records
Machine learning
- Forecasts enrolments, orders, stock and cash
- Spots the unusual in payments, payroll and stock
- Predicts no-shows, late payers and customers about to leave
- Plans rosters and delivery routes
Computer vision
- Reads receipts, invoices and paper forms
- Counts stock on a shelf from one photo
- Checks site photos for progress and safety gear
- Checks quality, like a bake’s colour or a finished wall
- Counts heads for ratio checks, without identifying anyone
Photos, video and anything you mark private are processed on the machine in your building, and never sent to an AI company.
| Job | Your AI computer | Cloud AI you approve |
|---|---|---|
| Replies to enquiries and messages | Yes | If you allow it |
| Payroll, salaries and bank details | Yes | Never |
| Reading receipts, invoices and forms | Yes | If you allow it |
| Forecasts and unusual-activity checks | Yes | Not needed |
| Checking photos and video | Yes | Never |
| Rare, very large reasoning jobs | If it has room | If you allow it |
If the machine is off, everyday work moves to the cloud AI you approved until it’s back. Local AI is part of Enterprise, and the hardware is bought in your name, at cost.
Questions about MCP
Do my staff need Claude, ChatGPT or Gemini?
No. Without MCP, everyone uses the AI chat built into CanopyBay. MCP is for staff who want to use the assistant they already know.
Does our data go to the AI company?
The records your AI reads pass through that AI’s provider, under your account with them. To keep everything in your building, use Local AI on Enterprise.
What does MCP cost?
MCP is included on Connected, at S$8 per staff, per module, a month, and on Enterprise. Your staff use their own AI plan, or the built-in chat.
What if an AI gets it wrong?
Nothing goes live until a person approves the preview, and every change can be undone.