Connect your AI assistant to Sage 50 Accounts.

Before you start

You need two things from your administrator:

1. The server URL — https://mcp.hyperext.com/mcp
2. Your sign-in credentials — an email address and password created for you

Your account is linked to your company's Sage 50 dataset. When your AI assistant connects, it can only see and change data belonging to your company.

How sign-in works

When you first connect, a browser window opens showing the HyperAccounts MCP sign-in page. Enter your email and password there — never give your password to the AI assistant itself. The sign-in has three quick steps:

1. Email + password on the sign-in page.
2. Set a new password — the first time you sign in (or after an administrator resets your password) you'll be asked to choose your own strong password. It needs at least 12 characters including an uppercase letter, a lowercase letter, a number and a symbol.
3. Email verification — we email a 6-digit code to your address. Enter it to authorise the connection. A new code is required each time you connect a new session, so keep access to your mailbox handy. Codes expire after 10 minutes; use Resend code if needed.

Connecting Claude

claude mcp add --transport http sage50 https://mcp.hyperext.com/mcp

Connecting ChatGPT

Custom MCP connectors are available on Plus, Pro, Team, Business and Enterprise plans (not Free). Full tool access (including creating/updating records) requires Developer Mode: enable it under Settings → Apps & Connectors → Advanced settings → Developer mode.

1. Open Settings → Connectors (via your avatar menu).
2. Click Create (New Connector).
3. Fill in:
   • Name: Sage 50 Accounts
   • MCP Server URL: https://mcp.hyperext.com/mcp
   • Authentication: OAuth
4. Leave the registration method as Dynamic Client Registration (DCR) — no client ID or secret is needed.
5. Click Create — your browser opens the HyperAccounts sign-in page. Enter your email and password and click Sign in & authorise.
6. In a chat, open the tools menu (or Developer Mode connector picker) and enable the Sage 50 Accounts connector for that conversation.

Connecting Gemini

The consumer Gemini app does not currently support custom MCP connectors — use Gemini CLI (or Gemini Code Assist, which shares the same configuration).

1. Install and run Gemini CLI at least once, then open the settings file:

   ~/.gemini/settings.json

2. Add (or merge) this section:

   {
     "mcpServers": {
       "sage50": {
         "httpUrl": "https://mcp.hyperext.com/mcp",
         "oauth": {
           "enabled": true
         }
       }
     }
   }

   Do not add a client ID, secret, or any tokens — Gemini discovers everything automatically and registers itself with the server.
3. Restart Gemini CLI, then authenticate:

   /mcp auth sage50

   Your browser opens the HyperAccounts sign-in page. Enter your email and password and click Sign in & authorise. Tokens are stored locally and refreshed automatically.
4. Verify the connection with /mcp — you should see sage50 listed with its tools.

What you can do

Once connected, your assistant has 100+ tools covering the full Sage 50 Accounts API:

Example prompts to try

• "Which customers have an outstanding balance over £1,000? Show their aged debt."
• "Find all sales invoices for account ABS001 from the last quarter and total them."
• "Create a customer called Northwind Traders, 12 King Street, Manchester M1 1AA, then read the record back to confirm."
• "Raise a sales order for CUST001: 5 × stock code S1 at the list price, deliver to their usual address."
• "What were our top 10 products by quantity despatched this month?"
• "Run the trial balance as at the end of last month and summarise the P&L lines."
• "Post a £250 bank receipt to nominal 4000 against account SMOK0001, dated today."

Tips for best results

• Dates are UK format dd/mm/yyyy almost everywhere. The one exception: the trial balance date uses dashes (dd-mm-yyyy) — your assistant is told this automatically.
• Be specific with account references (e.g. CUST001) when you know them; otherwise ask the assistant to search first.
• Ask for read-backs after writes — e.g. "…then read the record back to confirm". This makes the assistant verify its own work.
• The assistant can call usage_insights to see how a tool typically behaves (success rates, common errors, what's usually called next). Tool descriptions also improve over time as the server learns from real usage across all users — patterns only, never your data.
• Long tasks: searches return everything matching your filters — for big datasets, ask the assistant to filter by date range or account to keep responses fast.

Security & privacy

• The assistant acts as you — it can only access your company's data, with your account.
• Sign-in happens on the server's own page; your password is never shared with the AI vendor.
• Two-factor by email: every connection is confirmed with a one-time code sent to your email, so a stolen password alone cannot connect an assistant to your data.
• Strong passwords are enforced, and you set your own password on first sign-in — the administrator never knows it.
• Access tokens expire after 1 hour and renew automatically; your administrator can suspend your account at any time, which revokes access immediately.
• Every request and response is logged on the server for audit and support purposes.
• The cross-company "learning" feature shares only usage patterns (which tools are used, what errors occur) — never the contents of your accounting data.

Troubleshooting