Custom AI Connectors for Medical Practices: What 'Add to Claude' or 'Add to ChatGPT' Actually Means

Open Claude.ai's settings, or ChatGPT's, and you'll eventually find a screen mentioning "connectors" or "apps," with an "Add" button next to something that looks like a URL. There's no clear explanation of what that button does or why you'd want to press it. If that screen left you more confused than when you started, you're not missing something obvious — the concept just isn't explained well anywhere, and this article fixes that in plain language.

The short version: this is about your practice's own AI agents, built through ClinikEHR's Agent Studio, becoming something your everyday Claude or ChatGPT account can actually reach and use. Why that's worth understanding, in brief:

• One assistant, more tools — Claude or ChatGPT can gain the ability to check your schedule or billing status, instead of staying limited to general knowledge.
• No custom software project required — the mechanism is a standard both companies already support.
• You choose exactly what's exposed — a scheduling lookup, say, without handing over your whole system.
• Safety gates stay in place — nothing risky happens automatically just because a request came through a chat window.
• Works the same regardless of which assistant staff prefer — Claude and ChatGPT speak the same underlying standard.

Quick Answer

A "custom connector" (Claude.ai's term) and an "app" (ChatGPT's current term, after a December 2025 rename from "connectors") are two names for the same idea: telling your everyday AI assistant about an extra tool it's allowed to use, beyond what it already does by default. That extra tool is, under the hood, an MCP server — MCP stands for Model Context Protocol, an open standard letting an AI assistant discover and call a small "menu" of capabilities someone made available to it. For a medical practice using ClinikEHR, that menu is the practice's own Agent Studio agents — a scheduling lookup, a billing-status check — exposed through a dedicated endpoint the clinic controls. Add that endpoint to Claude.ai as a custom connector, or to ChatGPT as an app, and the assistant you already use can draw on your practice's own information, not just what it knew out of the box.

Why This Terminology Is So Confusing

Part of the confusion is simply that this is new. Most practice owners have used Claude or ChatGPT for drafting emails, summarizing documents, or brainstorming — none of which requires touching a settings menu labeled "connectors" or "apps." Most people encountering that menu are seeing the words used this way for the first time.

The rest comes from Claude and ChatGPT choosing different names for the same underlying idea. Claude.ai calls it a custom connector, found under Settings in a Customize or Connectors section, where a user adds any compliant remote server by pasting in its address. ChatGPT calls the broader feature apps — a name adopted in December 2025, replacing "connectors." Most ChatGPT apps come from a reviewed public directory and install in one click; for a server outside that directory — like a practice's own private one — ChatGPT offers a "Developer Mode" setting that adds it directly, the same way Claude.ai's screen does.

Two names, two slightly different menus, one identical concept underneath both. Once that's clear, the rest gets much simpler.

What a Custom Connector or App Actually Is

Strip away the branding and here's the plain idea: Claude and ChatGPT, by default, only do what they were built to do — talk, reason, write, search the internet if enabled. A custom connector or app tells one of them: there's an additional tool available, here's how to reach it, use it when relevant.

That additional tool is, technically, an MCP server — a small, standardized menu of things a piece of software lets an AI assistant do (check a schedule, look up a status), written in a format both Claude and ChatGPT already read, without custom integration code for each pairing. MCP stands for Model Context Protocol, an open standard that started at Anthropic and has since been adopted by OpenAI, Google DeepMind, and Microsoft — and as of December 2025, was donated to the Agentic AI Foundation under the Linux Foundation, making it vendor-neutral. See our companion explainer on what MCP is and how it works in healthcare AI for more.

Claude.ai's own support article on custom connectors and OpenAI's help center article on apps in ChatGPT both describe the same action: pointing the assistant at an MCP server's address.

How This Applies to a Medical Practice

Here's where it gets concrete. A practice using ClinikEHR's Agent Studio can build agents — a scheduling lookup, a billing-status check, an eligibility check — the same way it builds any other automated workflow. Agent Studio then lets the clinic expose chosen agents as an MCP server of its own: a dedicated endpoint, with its own address, offering only the agents it decided to include. Nothing is exposed by default.

That endpoint is what gets "added" to Claude.ai as a custom connector, or to ChatGPT as an app. From there, staff — using the account they already reach for out of habit — can ask it to check tomorrow's schedule or a billing status, and the assistant calls the practice's own agent to answer, instead of staying limited to general knowledge. No custom software gets built for this; the connecting logic is the open standard both platforms already support.

Two details matter here. Access can be handed over as a simple access token, best suited to developer tools and scripts, or a full sign-in using OAuth 2.1 — a modern login standard — built for Claude.ai's and ChatGPT's own connector and app screens, which expect a real sign-in rather than a pasted credential. Every endpoint also carries its own rate limit and usage cap, revocable independently at any time. For the exact click-by-click steps, see our step-by-step walkthrough — this article stays at the concept level rather than repeating them.

Challenges to Keep in Mind

This is a genuinely new capability across the industry, not a mature, settled feature. Claude.ai and ChatGPT are both actively iterating here; menu labels and even feature names have already shifted once (ChatGPT's "connectors" becoming "apps" is a recent example), and more change is likely. If a screen doesn't match a description you read somewhere, including this one, check the platform's current documentation rather than assume the capability disappeared.

Adding a connector also doesn't remove the need to think about what you're exposing. A scheduling lookup and something touching patient records directly carry very different risk, and being technically able to connect something isn't a reason to expose everything by default. Human-approval gates on genuinely risky actions — sending patient information, charging a card, submitting a claim — remain fully in force no matter which client made the request. Related reading: AI agents and HIPAA — what practices should know.

When This Is Worth Understanding and Using

This is worth your attention once two things are both true: you or your staff are already regular Claude or ChatGPT users, and there's something specific in your practice's own systems you keep wishing that assistant could just check without switching tools. If neither is true yet, there's no urgency. But if you recognize that itch, this concept explains what's actually possible — and the mechanism, an MCP server added as a connector or app, that makes it real without a custom development project.

Product Insight: ClinikEHR's Role in Custom Connectors

• Agent Studio lets a clinic create dedicated MCP endpoints, each exposing only the published agents it chooses — nothing is exposed automatically.
• Two connection modes: a one-time-shown access token for developer tools and scripts, and full OAuth 2.1 sign-in built for Claude.ai's and ChatGPT's own connector and app screens.
• Every endpoint carries its own independent rate limit and usage cap, and can be revoked or rotated without affecting any other connection.
• Patient data is de-identified and tokenized before it reaches an AI model, re-identified only inside ClinikEHR's own secure system.
• Human-approval gates on risky actions — sending PHI, charging a card, submitting a claim — apply regardless of which client the request came through.
• MCP-expose is available on any paid ClinikEHR plan, alongside the Free tier's 50 monthly agent credits.

See the Agent Studio feature page for the full picture, compare plans on pricing, and read the step-by-step connection walkthrough once you're ready to actually set one up.

Frequently Asked Questions (FAQs)

1. Is a "connector" the same thing as an "app"?

Yes, conceptually. Same underlying idea — adding an outside tool to Claude or ChatGPT — described with different names by different companies. Claude.ai calls it a custom connector; ChatGPT calls it an app, renamed from "connectors" in December 2025.

2. Do I need to know how to code to use this?

No, not to use an existing connector — adding one is mostly pasting in an address and signing in when prompted. Creating one from your practice's own agents involves some setup work, but ClinikEHR's Agent Studio handles that through a settings panel, not custom code.

3. What exactly is being "added" when I click Add?

Not a plug-in in the traditional sense — you're pointing Claude or ChatGPT at the address of an MCP server, a small standardized menu of capabilities. Once added, the assistant knows that menu exists and can call from it mid-conversation.

4. Is MCP a Claude-only or ChatGPT-only standard?

Neither. MCP originated at Anthropic but is now an open, vendor-neutral standard also adopted by OpenAI, Google DeepMind, and Microsoft, and was donated to the Agentic AI Foundation under the Linux Foundation in December 2025.

5. Does adding a connector give an AI assistant unrestricted access to my practice's data?

No. A clinic using Agent Studio chooses exactly which agents are exposed through an endpoint, and human-approval gates on risky actions — sending patient information, charging a card, submitting a claim — remain in force no matter which client is asking.

6. Where do I actually go to connect my practice's own agents?

This article explains the concept; the click-by-click steps are covered in our companion step-by-step walkthrough.

Conclusion

"Custom connector" and "app" are two names for the same simple idea: telling your everyday Claude or ChatGPT account about an extra tool it's allowed to use, in the standardized form of an MCP server. For a practice on ClinikEHR, that extra tool can be the practice's own agents — a scheduling lookup, a billing-status check — exposed through a dedicated endpoint the clinic controls, so the assistant staff already use can draw on the practice's own systems, safely and without a custom project.

Key takeaways:

• "Custom connector" (Claude.ai) and "app" (ChatGPT, renamed from "connectors" in December 2025) are the same concept with different names.
• The mechanism both rely on is MCP (Model Context Protocol), an open, vendor-neutral standard, not a proprietary feature of either company.
• Agent Studio lets a ClinikEHR practice expose chosen agents as its own MCP endpoint — the thing that actually gets "added."
• A clinic decides exactly what's exposed, with its own rate limits, usage caps, and revocable access per endpoint.
• Human-approval gates on risky actions apply no matter which assistant is making the request.
• Terminology and settings screens may keep shifting as both companies iterate — the underlying concept is the stable part.

See AI in action first with our Free Clinical Notes AI Generator — professional notes instantly, no signup, no credit card.

Ready to see what your own practice's agents could do inside the AI assistant you already use? Try ClinikEHR free, explore our pricing, or book a free demo.

Disclaimer: Claude.ai and ChatGPT are third-party products; their feature names, settings screens, and capabilities may change over time, as already happened when ChatGPT renamed "connectors" to "apps" in December 2025. This article is educational content, not a guarantee of compatibility with any specific version of a third-party product. ClinikEHR and its authors are not liable for decisions made based on this information.

Related Articles

• Can You Connect an EHR's AI Agents to Claude or ChatGPT? Here's How
• What Is MCP (Model Context Protocol) in Healthcare AI Agents?
• Agent-to-Agent (A2A) Connections for AI Agents
• Exposing an AI Agent as MCP — Safe for Patient Data?
• Dedicated MCP Endpoints: A Secure AI Connector
• How to Choose an AI Agent Platform for a Medical Practice
• AI Agents and HIPAA: What Practices Should Know
• MCP vs. Webhooks: How to Automate Your Practice