Athenahealth lets US patients read lab results with Claude for $20

Athenahealth is one of the larger EHR vendors in the US market, and its provider network covers practices serving roughly 20% of the American population, according to the company. The new component is small next to that footprint: a Model Context Protocol server that Claude calls when a patient asks something about their chart.

What the connector does

Model Context Protocol is Anthropic's open standard for connecting a model to outside data sources. It was published in November 2024 and has since been picked up by other vendors. An MCP server exposes a defined set of tools and data. Claude decides which one to call and then reads the result.

Here the patient authenticates with their normal athenahealth portal credentials. Queries then pull structured records on demand instead of pre-loading an entire file into the context window. Ten domains are exposed: lab results, vitals, medications, health conditions, immunizations, visit history, diagnostic reports, allergies, procedures and care plans.

On-demand retrieval keeps the prompt short, which affects latency and token cost. It also limits what the model sees to what the question actually needs.

What the data rules say

Athenahealth states that records fetched through the connector are not used to train Anthropic's foundation models. Patients can choose which categories to share and revoke access later. Those two points answer the first questions a privacy officer asks about any LLM integration in healthcare.

The commercial wrapper is a subscription. Claude Pro costs $20 a month, Claude Max $100 a month, and only paid tiers can enable the connector. EU customers are billed in EUR with local VAT added, so the European price sits above the USD list figure.

The arithmetic behind a $20 plan

Record questions are cheap at API rates. Take a modest user: 50 queries a month, each sending about 2,000 input tokens of retrieved record data plus a question, and returning roughly 500 output tokens. That works out to 100,000 input and 25,000 output tokens per month. Against Anthropic's published price list:

Ninety cents a month on Opus 5.5 at that volume. The paid subscription is not paying for these queries in isolation; it bundles them with everything else in the plan. A real connector session with several tool calls will usually push more context than my estimate assumes. The order of magnitude still explains why athenahealth can put the feature inside an existing subscription instead of billing per question.

The single-row table is deliberately narrow. The calculation uses Opus 5.5 because its published input and output prices are the figures cited here. A retrieval task like pulling ten lab values may well run on a smaller model, but that would only lower the token bill. Anthropic's price list is where the Opus figures come from.

Why European patients are not on the list

The connector is limited to US users on paid Claude tiers whose records sit on the athenahealth network. No European availability has been announced. A patient in Prague or Berlin with a Claude Pro subscription gets the same model and no path to their hospital record.

That gap is not only a product decision. Health data falls into a special category under Article 9 of the GDPR, which requires a legal basis such as explicit consent. In the athenahealth model the practice remains the data controller while the patient authorizes a transfer to a model provider at query time. Whether Anthropic acts as a processor, a separate controller or a recipient on the patient's instruction is the kind of question EU hospitals settle with their data protection officer before a pilot, not after.

The AI Act adds a layer that is already active. Since 2 August 2026, Article 50 requires that people be told when they interact with an AI system, and synthetic content must be marked. High-risk obligations for Annex III systems face a proposed postponement to 2 December 2027. With that change still proposed, a patient-facing query tool today is governed mainly by transparency rules, the GDPR and, where it qualifies, the Medical Devices Regulation.

Cross-border exchange is the other moving part. The European Health Data Space sets common rules for sharing health data between member states, and a consumer-facing connector would have to fit inside that framework rather than beside it.

What an EU deployment would need

An MCP server for European records would need consent that can be granted per category and withdrawn, an audit log of every record the model read, and a clear answer on where inference runs and how long logs are kept. None of that is exotic. All of it is legal and engineering work that a US rollout does not carry in the same form.

For developers the more useful part is the pattern. A provider exposes ten structured data domains through an MCP server, authenticates the user at the portal, and lets a general-purpose assistant handle the language. The protocol is open, so in principle the same server could front a different model. Anthropic gets a reference deployment inside a regulated industry. Athenahealth gets a feature whose inference cost is small change.

What it does not cover

The list of ten categories does not include imaging studies or free-text clinician notes, so those stay out of reach of the assistant. The launch material describes no way to ask about a family member's record or to hand a caregiver access. And there is no announced date for markets outside the United States.