Connect ChatGPT and Claude to Your Practice Data
Add CRMBridge as a connector, sign in once, choose a location, and your AI assistant can answer questions about the schedule, patients and production — and book appointments when you allow it to.
Ask Your Practice a Question
Most practice questions are simple to ask and tedious to answer. “How did hygiene production this quarter compare with last quarter?” “Which patients are overdue for recall and have nothing booked?” “Is there anything open with Dr. Patel on Thursday afternoon?” Each one usually means running a report, exporting it and filtering a spreadsheet.
ChatGPT and Claude can now answer them directly from the practice’s data. Both support connectors built on the Model Context Protocol (MCP), an open standard that lets an AI assistant use outside tools. The CRMBridge MCP server is one of those connectors: add it once, sign in with your CRMBridge account, choose which location the assistant may see, and it can read the schedule, patients, balances, production and recalls — across any of the 40+ practice systems CRMBridge supports.
Signing in through the browser means it works in the clients that cannot be configured with API keys: ChatGPT, claude.ai, Claude Desktop and Claude Code. Developers building their own agents can still connect with application keys.
Connecting Takes About Two Minutes
The connector address is the same everywhere:
https://mcp.crmbridge.ai/mcp
- ChatGPT: Settings → Connectors → Create, paste the address and choose OAuth as the authentication.
- claude.ai and Claude Desktop: Settings → Connectors → Add custom connector, then paste the address.
- Claude Code: run the command below, then type
/mcpto sign in.
claude mcp add --transport http crmbridge https://mcp.crmbridge.ai/mcp
Your browser then opens the CRMBridge developer portal. Sign in with your account and your second factor, and the consent page asks three things:
- Which app the assistant acts as. Calls are logged and billed against that app, exactly as if it used the app’s own key.
- Which location it may access. One location per connection; the assistant cannot see any other practice in your account.
- Whether it may make changes. Leave “Also allow changes” unticked for a read-only connection.
Click Allow and you are back in the chat. You stay signed in: access renews in the background, so you are not asked to sign in each time you use the connector. To disconnect, remove the connector in the client, or deactivate the app or the location in the developer portal; access stops within the hour.
What You Can Ask
The server gives the assistant 39 tools, all scoped to the location you chose. You do not call them yourself; you ask in plain language and the assistant picks the right ones. They fall into four groups.
Day-to-day lookups
Find a patient by name, phone or date of birth; see their next and last visit, balance and overdue recalls; list upcoming appointments; find open slots for a provider. For questions about the last few minutes, live tools read the practice’s own system at the moment you ask — a patient added this morning, a schedule change made at the front desk just now, the current account balance.
“Does Maria Lopez have anything booked, and what does she owe?”
“Show me tomorrow’s schedule for the hygiene chairs, including anything added today.”
Practice numbers
Analytics tools work through the whole period on the server and hand the assistant the totals: production and collections, kept, cancelled and no-show rates, treatment-plan acceptance and unscheduled treatment, recall compliance, open claims and accounts receivable aging, new and lapsed patients. Ask for a trend or a comparison and you get numbers you can check, grouped by month, provider, procedure or location.
“Compare production and collections by provider for Q3 against Q2.”
“What is our no-show rate by weekday over the last six months?”
Lists and cohorts
For exports and follow-up work: patients with no visit since a date, overdue recalls, accepted treatment without an appointment, open claims older than 60 days, the largest outstanding balances.
“List patients with accepted treatment over $1,000 and no future appointment, largest first.”
Changes, when you allow them
With changes allowed, the assistant can book, confirm, reschedule or cancel an appointment, register a new patient, update a patient’s contact details, add a note to the chart and attach a document. Each change goes straight into the practice system and the assistant reports back whether it was applied, refused (with the practice system’s reason) or still being confirmed.
“Find a 60-minute slot with Dr. Patel next week, afternoons only, and book the first one for James Chen.”
Keeping Changes Under Control
- Read-only is one checkbox away. A connection made without “Also allow changes” cannot write anything; a write attempt is refused.
- The client asks first. Every write tool is marked as one that changes data and every read tool as read-only, so the client asks for your approval before a change runs.
- No accidental double bookings. If the assistant calls the same write twice within ten minutes — a retry after a timeout, or the model repeating itself — the second call returns the first result instead of writing again.
- Nothing is deleted. Appointment changes confirm, change status or reschedule; no tool deletes records.
- Every call is accounted for. Calls are logged against the app you chose, so the API dashboard shows how many calls the assistant made and to which endpoints.
Patient Data and the AI Provider
When the assistant reads a patient’s record, that information is sent to the AI provider you are using so it can answer. Treat the connection like any other system that handles protected health information.
- Use an AI plan and provider whose agreement covers health data where your rules require one (for example a business associate agreement under HIPAA), and follow your local privacy law.
- For exploring and demos, connect a location that runs on a demo or test copy of the practice system rather than a live practice.
- Choose the narrowest connection that does the job: one location, read-only unless changes are needed.
- Review the connections you have granted from time to time, and remove the ones you no longer use.
For Developers
The MCP server is an AI-facing layer over the same REST API, so anything an assistant can do through it, your own code can do through the API. Use the REST API for server-to-server integrations, and the MCP server when an AI agent should call CRMBridge directly. Clients that take a configuration file, such as Cursor or your own agent, can connect with your application key instead of signing in; the MCP documentation lists every tool, the headers and an example configuration.
A good first session
Connect read-only, ask for last month’s practice dashboard, then ask follow-up questions about anything that surprises you: which provider, which weeks, which procedures. It is a quick way to learn what the tools can do before you build anything.
Connect Your Assistant
Create a developer account, link a location, and add https://mcp.crmbridge.ai/mcp as a connector in ChatGPT or Claude.