I’m building HealthPorta — an API layer for the US healthcare data (providers, plans, coverage, costs). Like most healthtech APIs, it started small… and then it didn’t. The surface area grew fast, integrations took longer than they should, and every new partner ended up needing a mini “tour guide” to use it correctly.
So, since the last post here, something changed. We’ve built an MCP server as a boring integration helper to provide a smaller, friendlier interface for engineers of our "two clients" who didn’t want to learn the entire API on day one and just wanted to ask Codex or Claude to "integrate".
Then something unexpected happened.
Once it was hooked into MCP clients, it stopped feeling like "API integration tooling" and started feeling like something you can talk to:
- Find a pcp doctor near me - Does this plan cover this medication? - What’s the cost signal for this procedure/provider? - Any quality or cost indicators for this provider?
Under the hood, it’s still structured tool calls — but the interaction flips from reading the docs & mapping endpoints to "ask a question + inspect the returned structure." Frankly speaking, it was a good stress test for the API layer, as, by common consensus, the MCP tested the APIs quite well, from the structure of calls to usage limits also.
What it can help with (today):
- provider lookup (specialty/location) - plan & issuer lookup - medication details, including restrictions, etc. - coverage/formulary-style checks (where available) - cost-related signals + basic comparisons (doctors, plans) - provider metadata + quality-related signals (where available)
What we’re working on now: We’re going deeper on transparency data from insurers and hospitals to improve cost/coverage accuracy and make outputs more explainable (and less "trust me, bro").
There’s a free per-user cap, so you can test it without paying or adding a card. Use it from your favorite MCP-capable console or IDE (anything that supports MCP and auth).
I’d love feedback on it.
Happy to answer questions / share implementation details.
endurant_dev•1h ago
``` ...updated the ranking for knee MRI providers (CPT 73721) relative to ZIP 60654, including their cost levels, average submitted charges, quality scores, and in-network status for both BCBS IL PPO and HMO plans.
```While it takes series of calls + analysis, the total time for this request is about 50 seconds from gemini-cli. It does requests for plans, doctors' information + of course Gemini client wants to use Google search to make it even better(for us this is an indicator of point of data we need to add), as it did a comparison of plan data (benefits) with the condition of the patient:
> Client, a family of 2+2, moves to Miami, we need to help them select a plan for 2026, while the daughter (13 y.o.) often has an illness (ears). One of the parents has not-healthy liver
``` I have analyzed the health plan options for Miami in 2026, focusing on the family's specific medical needs. For a family with chronic conditions (pediatric ear issues and liver health), a Platinum or Gold plan is highly recommended due to lower deductibles and comprehensive specialist coverage.
```The main focus on mcp tools for end-user (integration with ChatGPT as an app or anything else) requires not only batching the requests (just to speed up), but cleaning up the responses from the data or structures that the client doesn't need (based on the request, which normal APIs can't do) to fight with tokens overusage for nothing and providing the most efficient plan/scenario for the requesting the data the client needs.