Connect Your Health Records to Claude (and ChatGPT) With MCP

Yes — with the right connection, Claude can answer questions about your real blood test results, pulling your actual values instead of guessing about a generic person. The difference between "what's a healthy ApoB?" and "how has my ApoB moved since March?" is whether the AI can see your data. Libby's integration with the Model Context Protocol (MCP) is what closes that gap for Claude: it lets the assistant read your lab history directly and read-only, so a question about your trend gets an answer grounded in your record. This guide explains what MCP is in plain terms, what the connection does, how to set it up at a high level, and — most importantly — how you stay in control of what's shared.

None of this replaces a clinician. What it does is let you ask sharper questions against a longer memory than any single appointment provides.

Can Claude read my blood test results?

Claude can read the results it's given. On its own, it can't reach into a lab portal or know your history — like any assistant, it only knows what you paste, upload, or connect. You have two good ways to hand it your labs:

• Paste or upload a report or a structured summary into the chat. Fast, works anywhere, but it's a one-shot snapshot — you're re-supplying context every session. (Our companion guide on using lab results with ChatGPT and Claude covers this path in depth.)
• Connect your record so Claude can query it directly. This is what MCP enables — instead of you pasting numbers, the assistant fetches the exact values it needs from a source you control, read-only.

The second option is the one that finally earns the phrase "analyze my labs," because the model can see every reading of every marker at once rather than a lonely dot you happened to paste.

What is MCP, in plain terms?

MCP — the Model Context Protocol — is an open standard (originally developed by Anthropic) that gives AI assistants a common way to plug into outside tools and data.¹ Think of it as a universal adapter — its own docs compare it to a USB-C port for AI.² Before a standard like this, every app needed its own bespoke bridge to every AI; with MCP, a service can expose a small set of well-defined "tools" that any MCP-aware assistant knows how to call.

It's worth knowing that MCP is quickly becoming the shared standard for this, rather than a one-off Libby trick: it's an open protocol supported across a range of assistants and tools — Claude and ChatGPT among them — so "connect your data" is turning into something you can expect to do the same way everywhere, not a proprietary bridge.² Other tools reasonably take a "paste your export" approach, and that works fine; a live connection is just a different tradeoff, covered below.

How a live connection differs from "paste your data"

Both get your numbers to the AI, so the difference is in how:

• Pasting an export hands over a static snapshot. It's fast and works anywhere, but you re-supply it every session, it's as current as the moment you copied it, and that text then lives in the chat's history.
• A live MCP connection lets the assistant run a scoped, read-only query against a source you control, and it returns just the values asked for — your own numbers against your lab's reference ranges — pulled fresh from your record rather than a copy you pasted. You can switch it off whenever you want.

Neither is "the right one" for everyone — a quick paste is perfect for a one-off. But for anything you'll ask more than once, a scoped connection means you're not repeatedly pasting a document you'd rather not hand over whole. (Our companion guide on whether it's safe to upload health records to ChatGPT digs into that tradeoff.)

Libby runs one of these connections. When you enable it, Claude gains a set of read-only tools — "list this person's markers," "get the history for ApoB," "pull this report" — and can call them to answer your questions. It reads your data; it never changes it. And it returns your lab's reference ranges, not invented "optimal" targets — so what counts as optimal versus merely in range stays a judgment for you and a clinician, not a number the tool hands down.

You don't need to be a developer to use it. MCP is the plumbing; the experience is just asking Claude a question and getting an answer that knows your numbers.

Why AI plus your real history beats generic advice

Ask any AI "what's a normal cholesterol level?" and you get the population answer: reference ranges, general caveats, see your doctor. It's fine, and it's the same answer everyone gets. The useful signal in lab work lives in two places a generic question can't reach:

• Your context. Reference ranges shift with age and sex, and "in range" isn't the same as "optimal". An AI can only account for that if it knows who you are.
• Your trend. A single ApoB, HbA1c, LDL, or Lp(a) value is one morning's dot. The story is the direction across draws — "ApoB down 22% since I changed my diet in March" says far more than "ApoB is 73."

A connected record feeds both in automatically. That's the whole point of wiring an AI to your longitudinal data instead of a one-shot upload: the model stops reasoning about a generic person and starts reasoning about you.

What Libby's MCP connection lets you do

Once your Libby record is connected, you can ask Claude the kinds of questions that only make sense against a real history:

• "How has my ApoB moved over the last three years?" — Claude pulls each reading from your record and describes the trajectory, instead of you transcribing numbers into the chat.
• "Which of my markers are trending the wrong way?" — because the assistant can see every marker on one timeline, not a single pasted value.
• "Summarize my last panel and the ranges each result fell in" — grounded in the actual reference range printed on your report.

The setup that makes this work is Libby's day job: drop in your lab PDFs from Quest, LabCorp, Function, BostonHeart, or a hospital on Epic or Cerner, and Libby reads every value, reconciles the units, and files each marker onto one timeline. That reconciled history is exactly what an AI needs to be useful — and it powers your biological age (PhenoAge), computed from your labs, too. The MCP connection just hands Claude a read-only window onto it.

If you'd rather not connect anything, you can still generate a clean, structured summary and paste it into ChatGPT or Claude by hand — same idea, more manual. And because it's a record you own, the AI conversation is one use of your data, never the place it's trapped.

How to connect your Libby record to Claude

The exact buttons evolve as the apps update, so treat this as the reliable path rather than a pixel-perfect map:

1. Build your record first. Start your record and import a lab PDF or two, so there's a history worth querying. The first upload takes about a minute.
2. Turn on the connection in Libby. Enable the Claude/MCP integration from your Libby account — this is where you opt in and control access.
3. Add Libby in Claude. Claude discovers Libby as an available connector; once it's added, Libby's read-only tools show up for the assistant to use.
4. Ask your questions. Now "how has my ApoB moved since March?" is answered from your real values. You can revoke the connection whenever you want.

For ChatGPT specifically, the fastest reliable path today is still to share a structured export of your record — see the ChatGPT and Claude lab-results guide for how to do that well, and how to export and keep your data either way.

Who controls your data?

You do — and that's the honest, verifiable claim we lead with. A few points worth being precise about:

• You choose what to share. The connection is something you turn on, and you can turn it off. Sharing your data with an AI is a decision you make, not a door you can't close.
• It's read-only. The MCP tools read your record to answer questions. They don't write to it, and they don't change your results.
• You own and export your data — no lock-in. Your record is yours to take with you, for life. That's the point of keeping the source of truth in a record you own rather than trapped inside a chatbot's history.

We're deliberately not going to make security claims we can't stand behind here. What we can say plainly is that the goal is for your data to stay yours, so that handing it to an AI is a choice on your terms.

What's live vs. what's coming

Being straight about scope matters, so here's the line:

Live today: import lab-result PDFs (Quest, LabCorp, Function, BostonHeart, Epic/Cerner), every marker reconciled onto one timeline, biological age (PhenoAge) computed from your labs, sharing with your clinician or with Claude via the MCP integration, and full ownership and export of your data.

On the roadmap (not yet live): connected/automatic portal import (today it's PDF import, not auto-sync), wearables (Oura, Whoop, Apple Health, CPAP), and family records. We'll say when those ship — we won't imply they're here before they are.

If you want your labs in a form an AI can actually reason over, start your record and import your first PDF — then decide, on your terms, what to share.

FAQ

Can Claude read my blood test results? Yes, if you give it access. Claude can read a report you paste or upload, or — via Libby's MCP connection — query your lab history directly and read-only, so it reasons over your real trend instead of a single pasted number. It can't reach into a lab portal on its own.

What is MCP? The Model Context Protocol is an open standard (originally developed by Anthropic) that lets AI assistants plug into outside tools and data through a common interface. Libby uses it to give Claude a read-only window onto your lab history.

Do I need to be a developer to connect Libby to Claude? No. MCP is the underlying plumbing; using it is just enabling the connection in Libby, adding Libby in Claude, and then asking questions in plain language.

Does connecting my labs to an AI change or store my results somewhere new? The connection is read-only — it reads your Libby record to answer questions and doesn't alter it. Your data stays in a record you own and can export at any time.

Can I do this with ChatGPT too? You can use your Libby data with ChatGPT — the most reliable path today is sharing a structured export of your record. See our lab-results-with-ChatGPT-and-Claude guide for the step-by-step, and whether it's safe to upload health records to ChatGPT for the privacy tradeoffs.

---

Educational content, not medical advice. This article is for general information and personal record-keeping. AI tools can make mistakes and are not a diagnosis or a treatment plan, and a lab reference range is not the same as an "optimal" target — it varies by lab and by person. Always talk to a qualified healthcare professional about your results and any decisions that follow from them.

Footnotes

1. Introducing the Model Context Protocol — Anthropic. Anthropic's announcement of MCP as "a universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol." ↩

2. What is the Model Context Protocol (MCP)? — modelcontextprotocol.io. The protocol's own documentation: MCP is an open-source standard for connecting AI applications like Claude or ChatGPT to external systems, likened to "a USB-C port for AI applications," and supported across a wide range of clients and tools. ↩ ↩²