Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Medplum

Open-source FHIR server, auth and React kit for healthcare apps, with an LLM operation and an MCP endpoint built in.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

4.0/5

A serious, standards-based backend for healthcare apps that makes LLM features easy to wire in, while leaving the hard compliance questions about sending patient data to a model provider with you.

What is Medplum?

Medplum is an open-source developer platform for healthcare applications built on FHIR, the standard data format for health records. It bundles a FHIR clinical data repository on PostgreSQL, OAuth and SMART-on-FHIR authentication, server-side Bots for custom logic, TypeScript SDKs and a React component library.

Its AI features are small but useful. The $ai operation passes chat messages and FHIR tool definitions to OpenAI, or to any OpenAI-compatible proxy such as LiteLLM, using a key kept in project secrets. Spaces, an assistant in the Medplum Provider app, chains three Bots to turn questions into FHIR queries, summaries and charts. A hosted MCP server exposes a fhir-request tool to clients such as Claude.

The repo has about 2,800 stars and an Apache-2.0 licence, and it ships often: four releases came out in the first 11 days of November 2026, the latest v5.2.17.

Is Medplum worth it?

Medplum is a FHIR platform first and an AI one second, which is the right order. Health data needs access policies, audit trails and a standard schema before any model touches it, and Medplum supplies those. Its docs describe how Spaces works: the browser, not the model, executes the FHIR requests the model proposes, under the signed-in user's AccessPolicy, so the assistant cannot read or write anything that user could not.

The AI pieces are simple by design. $ai is a FHIR operation that forwards an OpenAI-format conversation, optional tools and a reasoning-effort setting to OpenAI by default, or to the URL in an LLM_BASE_URL secret. Patients cannot call it directly; patient-facing features must go through a Bot. The MCP server takes the opposite approach to narrow tools: one fhir-request tool with full create, read, update and delete rights, on the reasoning that models already know FHIR well. That is powerful, and most users should get read-only access policies with it.

The catch: the $ai documentation says nothing about protected health information or a business associate agreement with OpenAI. Sending patient records to a model provider under HIPAA needs that agreement, and it is your job to have one or to point LLM_BASE_URL at a model you host. Spaces also needs the Medplum team to switch it on, and its write actions, such as orders and referrals, happen as soon as the loop reaches them.

Cost is the other consideration. The hosted Free plan is for prototyping; Production starts at $2,000 a month and Premium at $6,000, and among paid plans only Enterprise includes a self-hosted option. You can run the Apache-2.0 code yourself on Ubuntu with PostgreSQL and Valkey, but Medplum's own guidance says self-hosting at scale needs skilled SRE staff, with a single upgrade taking one week to one month of an engineer's time. About 780 open issues and pull requests reflect a large, busy codebase.

Where it earns its place

  • A five-engineer digital-health startup

    Ship a patient portal and clinician app without writing a FHIR server

    Prototype on the hosted Free plan with the React components and TypeScript SDK, then move to Production, from $2,000 a month, once real patient data and an MSA/BAA with Medplum are needed.

  • A clinical informatics team at a multi-site clinic

    Let clinicians ask for patient summaries and lab trends in plain English

    Have Medplum enable Spaces, write the three system-prompt Communications, and give users read-only AccessPolicies so the assistant can search and chart but not place orders. Model costs go to the OpenAI or proxy account in project secrets.

  • A health-data integration engineer

    Query a FHIR store from Claude while building integrations

    Add the hosted MCP endpoint as an organisation integration on a paid Claude plan, sign in to Medplum and use fhir-request against a test project, keeping production data out until access policies are tightened.

Before you commit

Setup
Sign up for the hosted Free plan, or self-host on Ubuntu 24.04 with at least 4 GB of RAM, PostgreSQL 16, Valkey and Node.js from Medplum's APT repository; Docker, Kubernetes, AWS, Azure and GCP guides also exist.
Integrations
FHIR REST and GraphQL APIs, TypeScript SDK, React components and SMART-on-FHIR; $ai to OpenAI or OpenAI-compatible endpoints; MCP for AI clients; AWS Textract and Comprehend Medical for document OCR.
Data and privacy
Medplum Cloud holds ONC Health IT certification and offers an MSA/BAA; the open-source code itself carries no certification. $ai sends message content to the configured model provider with a key from project secrets.
Cost at scale
Production starts at $2,000 a month and Premium at $6,000, plus usage fees beyond included allowances for e-prescribing, labs, fax and clearinghouse; Enterprise and self-hosting are by quote.
Limits
Tool calls are not supported in $ai streaming mode; the Spaces loop runs up to 10 iterations; managed UMLS terminology is hosted-only; about 780 open issues and pull requests.

What’s good

  • Standard FHIR data model, with access policies applied to AI-proposed requests
  • $ai works with OpenAI or any OpenAI-compatible proxy, including LiteLLM
  • Hosted MCP endpoint for Claude and other clients
  • Apache-2.0 licence and a very active release cadence

What’s not

  • No published guidance on PHI or BAAs for the $ai model provider
  • Paid hosting starts at $2,000 a month
  • Spaces needs the Medplum team to enable it
  • Self-hosting at scale demands dedicated SRE time

Best for

Engineering teams building a clinical or digital-health product who want a FHIR backend with LLM hooks and will handle their own model-provider compliance.

Medplum: common questions

Is Medplum worth it?
A serious, standards-based backend for healthcare apps that makes LLM features easy to wire in, while leaving the hard compliance questions about sending patient data to a model provider with you. We rate it 4.0 out of 5.
How much does Medplum cost?
Apache-2.0 code free to self-host. Medplum Cloud: Free for prototyping; Production from $2,000/month; Premium from $6,000/month; Enterprise by quote, with a self-hosted option. Usage above included allowances is billed separately.
What are the downsides of Medplum?
No published guidance on PHI or BAAs for the $ai model provider; paid hosting starts at $2,000 a month; spaces needs the Medplum team to enable it; self-hosting at scale demands dedicated SRE time.
Who is Medplum best for?
Engineering teams building a clinical or digital-health product who want a FHIR backend with LLM hooks and will handle their own model-provider compliance.
How much does Medplum cost for a team or in production?
Production starts at $2,000 a month and Premium at $6,000, plus usage fees beyond included allowances for e-prescribing, labs, fax and clearinghouse; Enterprise and self-hosting are by quote.
Is Medplum safe for company data?
Medplum Cloud holds ONC Health IT certification and offers an MSA/BAA; the open-source code itself carries no certification. $ai sends message content to the configured model provider with a key from project secrets.

Answers checked Nov 16, 2026.

Make Medplum? Show this rating on your siteMedplum rated 4.0 out of 5 on ReviewRamble