An MVP under MDR is not a shippable clinical product. It is the smallest build that lets you test a hypothesis without placing a device on the market. Anything sold, distributed, or used on patients as a medical device must be CE-marked. Everything before that point lives in research, investigation, or internal prototyping — with strict rules about what you can and cannot do.

By Tibor Zechmeister and Felix Lenhard.

TL;DR

  • The SaaS definition of MVP does not apply in MedTech. You cannot "launch and iterate" a medical device with patients.
  • Under MDR Article 5, no device may be placed on the market or put into service unless it complies with the regulation, which for most devices means CE marking.
  • Article 21 allows devices intended for clinical investigation to move without CE marking, but only inside the strict regime of Article 62 and Annex XV.
  • A real MedTech MVP tests a specific hypothesis — usability, workflow fit, clinical signal — without crossing the line into "placing on the market".
  • Intended purpose (Article 2(12)) should stay as narrow as possible during MVP work, because every word eventually becomes a claim your notified body will audit.
  • The right question is not "how do we ship faster" but "which hypothesis, if wrong, kills the company — and what is the smallest regulated-safe experiment that answers it".

Why this matters

Most MedTech founders come from software, consumer, or hardware backgrounds where the MVP is sacred. Build the smallest thing, get it in front of users, learn, iterate. Paul Graham's line — "launch something you're embarrassed by" — is etched into startup culture.

Apply that logic to a medical device and you break the law. Worse, you can harm patients and end your company before it starts.

But the opposite extreme is also wrong. Founders who take "MVP is dead in MedTech" literally end up building a gold-plated Class IIb device for three years against a clinical problem they never validated. They raise a Series A, hit the market, and discover that clinicians do not want the workflow they designed. That is not regulatory caution. That is waste dressed up as rigour.

The real answer is in the middle, and it is precise. There is a form of MVP that works under MDR. It does not look like a SaaS beta, but it also does not require a full technical file on day one. Understanding which experiments you can run — and under which legal regime — is what separates founders who ship in 18 months from founders who spend 18 months confused.

What MDR actually says

Three provisions matter most for MVP thinking.

Article 5 (Placing on the market and putting into service). Paragraph 1 is unambiguous: a device may only be placed on the market or put into service if, when duly supplied and properly installed, maintained, and used in accordance with its intended purpose, it complies with the regulation. In practice that means a device you intend to be a medical device cannot be sold, given away, or deployed for clinical use in the EU without CE marking (or a specific exemption).

Article 2(12) (Intended purpose). "Intended purpose means the use for which a device is intended according to the data supplied by the manufacturer on the label, in the instructions for use or in promotional or sales materials or statements and as specified by the manufacturer in the clinical evaluation."

The intended purpose is not what you privately think the device is for. It is whatever you say to the market — on your website, in a pitch deck, in a user manual, in the product UI. Every claim expands or contracts your regulated scope.

Article 21 (Devices for special purposes) combined with Article 62 and Annex XV. Devices intended for clinical investigation may be supplied to investigators without CE marking, provided the investigation is conducted under the rules of Chapter VI. This is a genuine MVP corridor — but it is narrow, formal, and requires an approved clinical investigation plan, ethics committee sign-off, sponsor obligations, insurance, and competent-authority notification.

There is no general "we're just testing" exemption. "Test market", "pilot deployment", and "friendly-user beta" are not legal categories under MDR.

A worked example

A three-person startup is building an AI-based triage tool for dermatology. The founders believe the product is medical-grade Software as a Medical Device. Before they spend 18 months on a Class IIa certification, they need to validate three things: whether dermatologists trust algorithmic triage at all, whether the algorithm performs on real clinical images, and whether the UI fits into a clinic workflow.

A SaaS-style MVP would release a web app tomorrow and let five dermatology clinics "try it". That route is illegal. The moment the tool outputs a triage recommendation for a real patient in a clinic, it is being put into service as an unregistered medical device, and Article 5 is breached.

Here is a lawful MVP plan that tests all three hypotheses:

  1. Workflow fit — no device at all. Build a figma prototype. Run moderated usability sessions with 12 dermatologists using synthetic cases. No real patient data, no clinical output, no claim of medical utility. This is product research. MDR does not regulate Figma prototypes.

  2. Algorithm performance — bench evaluation on retrospective data. Buy or license a de-identified image dataset. Run the algorithm offline. Compare outputs to ground-truth labels. This is non-clinical performance evaluation and contributes to the future technical file. No device is placed on the market because no clinician is using the output to make a decision.

  3. Clinical trust and real-world performance — clinical investigation under Article 62. If, after steps 1 and 2, the signal is strong enough to justify the cost, design a small clinical investigation under Annex XV. Get ethics approval. Notify the competent authority. Supply the investigational device to three sites under Article 21. Now you are legally using the device on patients — inside the investigation regime, not on the open market.

Notice what happened. The team tested three hypotheses in three different legal regimes, in increasing order of cost and seriousness. They never "just released" a medical device. They also never built a full Class IIa dossier to answer a figma-level question.

This is what an MVP means under MDR. Not a smaller product. A smaller experiment, run in the correct legal regime.

The Subtract to Ship playbook

Here is the playbook we use with startups to keep MVP work honest under MDR.

Step 1: Write the intended purpose last, not first. A lot of founders rush to draft a one-paragraph intended purpose in month one and then spend the next year defending every word to their notified body. Do the opposite. During early MVP work, write your intended purpose as a loose working hypothesis marked "draft — not for external use". Refine it with every round of evidence. Lock it only when you commit to the full technical file. See our guide on defining intended purpose without over-constraining.

Step 2: Separate the three regimes cleanly. At any moment, every experiment you run should be sortable into one of three buckets: non-device research (no patient, no medical claim), bench/non-clinical evaluation (no patient interaction), or formal clinical investigation under Chapter VI. If you cannot name the regime a specific experiment lives in, stop and name it before you run it.

Step 3: Make a written list of things you will not claim yet. Not on the website, not in the pitch deck, not in the demo script. "Diagnoses", "detects", "recommends treatment", "reduces risk of" — every one of these is a medical claim that can push you into device territory under Article 2 and Annex VIII Rule 11. Our post on wellness-first, medical device later unpacks how founders sometimes use a non-medical positioning to buy time legally.

Step 4: Build the QMS kernel in parallel, not later. You do not need full EN ISO 13485:2016+A11:2021 certification to run an MVP experiment. But you do need a lightweight design and risk management system from the first line of code you intend to keep, because retrofitting a design history file after the fact is the single most expensive mistake we see. A two-phase development approach lets you run lean in phase 1 and scale the QMS into phase 2 once the hypothesis is proven.

Step 5: Budget one clinical investigation into the MVP phase, not into the launch phase. Founders consistently under-budget the Article 62 pathway. If your hypothesis genuinely requires patient exposure to be tested, the clinical investigation is the MVP. It is not something that happens "later, once we have product-market fit". The investigation is how you get product-market fit.

Step 6: Validate before regulate. Everything that can be learned without touching a patient should be learned before touching a patient. That includes willingness-to-pay, workflow integration, reimbursement pathway, decision-making unit dynamics, and market size. Our guide on validating a MedTech idea before MDR covers the pre-regulatory discovery work in detail. And our piece on the minimum viable regulatory strategy shows how to scope CE work to what is genuinely needed.

Reality Check

  1. Can you state, in one sentence each, the three hypotheses your current MVP is designed to test?
  2. For each experiment you are running this month, which legal regime is it in — non-device, non-clinical evaluation, or Article 62 clinical investigation?
  3. Is there anything on your website, in your pitch deck, or in your demo that makes a medical claim you have not substantiated?
  4. Is your draft intended purpose locked or still a working hypothesis? Do you have it in writing which?
  5. If a competent authority inspector walked into your office tomorrow, could you explain how your current activities comply with Article 5?
  6. Have you written down the design and risk records from today's work, or will you be reconstructing them in year two?
  7. If the clinical investigation you will eventually need costs 400,000 EUR and takes 14 months, is that on your roadmap — or is it still "later"?

Frequently Asked Questions

Can we run a "friendly-user beta" with a few clinics before CE marking? Not if the device is being used for its intended medical purpose on real patients. That is placing on the market or putting into service under Article 5. The only legal route to put an uncertified device in front of patients is a clinical investigation under Article 62.

Is a bench prototype regulated by MDR? A prototype that never interacts with a patient and is not offered for sale is not placed on the market. Internal testing and non-clinical performance evaluation are unregulated as such, though they should still be documented in your design history file to be reusable later.

What about a research-use-only label? MDR has no "research use only" exemption for devices intended to become medical devices in the EU. That label exists in other jurisdictions and for in-vitro contexts. Do not rely on it as an MDR workaround.

Can we pivot the intended purpose after we start? Yes, and you should, as long as you are still inside the experimentation phase and nothing has been placed on the market. Every pivot must be reflected in your draft documentation and risk analysis. Once you submit for conformity assessment, changes become significantly more expensive.

Does MVP thinking apply at all to Class III devices? Yes, but the non-clinical phase is longer and the clinical investigation is larger. The principle is the same: test the riskiest hypothesis in the cheapest legal regime that can answer it.

What if we launch a non-medical version first? That is a legitimate strategy for some products and a misleading shortcut for others. It depends on whether the non-medical version can stand on its own commercially and whether your marketing avoids medical claims entirely.

Sources

  1. Regulation (EU) 2017/745 on medical devices, consolidated text. Article 2(12), Article 5, Article 21, Article 62, Annex XV.
  2. EN ISO 13485:2016+A11:2021 — Medical devices — Quality management systems — Requirements for regulatory purposes.
  3. EN ISO 14155:2020+A11:2024 — Clinical investigation of medical devices for human subjects — Good clinical practice.