NPI MCP Server
NPI MCP Server connects AI clients to US provider registry data through the Model Context Protocol. Billing and credentialing teams ask in plain English and get structured answers.
- MCP server
A Model Context Protocol server that lets an AI client look up, search and validate US healthcare provider NPI registry records.
- Model Context Protocol
- Server-sent events (SSE) transport
- HTTP endpoints
- CMS NPPES provider data
- OIG LEIE exclusion list
- NUCC provider taxonomy codes
- Claude
- Cursor
- Windsurf
- Continue
The challenge
Medical coders, billing teams and credentialing staff check provider identifiers all day. A wrong or mistyped NPI, a mismatched specialty or an outdated address can send a claim back, and each check usually means a separate lookup site.
Compliance adds another step: confirming that a provider or organisation is not on the federal exclusion list before any claim is submitted. These checks are repetitive, easy to skip under time pressure, and hard to run inside the AI tools teams already use.
What it does
NPI lookup
Retrieve the registry record for a given NPI, including provider name, individual or organisation type, specialty with a plain-language description, practice location and phone number.
NPI validation
Check that an NPI is well formed and passes the check digit test, which helps catch typing errors before a claim is submitted.
Provider search
Find providers by name, state, city or ZIP code, with wildcard support for partial names and ZIP prefixes, and paging to move through longer result lists.
Specialty search
Search by plain-English specialty such as cardiology or family medicine, or by exact taxonomy code. Descriptions are translated to matching taxonomy codes automatically.
Organisation search
Find hospitals, clinics and other organisation records by name fragment, state or ZIP code, separate from individual practitioners.
Exclusion check
Check a provider against the federal OIG exclusion list by NPI, by name with date of birth, or by organisation name, with a pointer to the official site for final confirmation.
How it works
Connect a client
Add the server to an MCP-compatible AI client, such as Claude, Cursor, Windsurf or Continue, using its server address and SSE transport.
Ask in plain English
Type a request such as a lookup, a validation, a specialty search or an exclusion check. The client selects the matching server tool.
Query the registry data
The server searches the NPPES provider data and, for exclusion checks, the OIG exclusion list, both loaded on a regular refresh schedule.
Review and confirm
The client presents structured results, including status, match type and exclusion details. The reviewer confirms important hits at the official source before acting.
Under the hood
- Protocol
- Model Context Protocol
- Server-sent events (SSE) transport
- HTTP endpoints
- Data sources
- CMS NPPES provider data
- OIG LEIE exclusion list
- NUCC provider taxonomy codes
- Clients
- Claude
- Cursor
- Windsurf
- Continue
- Formats
- JSON configuration
- CSV exclusion file
- Structured text responses
Inside the harness
Policy
Decide which teams may connect an AI client to this server and which request types, such as exclusion checks, are allowed for each role.
Security
The reference setup needs no login. In a governed deployment, place the server behind authenticated access and review what provider data leaves the network.
Observability
Track which tools are called, how often and how long they take, so unusual query volumes or failing lookups are visible to operators.
Audit
Record each lookup and exclusion check with who asked and when, giving billing and compliance teams a trail for their verification steps.
Business outcome
Fewer avoidable claim errors. Validating identifiers and confirming specialty and location before submission helps teams catch simple mistakes earlier, before a claim is rejected.
One place for routine checks. Lookups, searches and exclusion checks run from the AI client staff already use, replacing a set of separate sites and manual steps.
Exclusion screening built into the flow. Teams can check providers and organisations against the federal exclusion list as part of ordinary verification instead of as a separate, easily skipped task.
Open starting point for integration. Because the work follows an open protocol, engineering teams can adapt the same pattern to their own data sources and internal controls.
Where it is used
- A medical coder validates an NPI and confirms the listed specialty before submitting a claim.
- A credentialing team searches for providers by specialty and ZIP prefix to build a network list.
- A compliance officer checks a new vendor or provider against the federal exclusion list during onboarding.
- A developer connects the server to an internal AI assistant as a working example of a healthcare data tool.
Frequently asked questions
How do I use it?
Add the server address to an MCP-compatible client such as Claude, Cursor, Windsurf or Continue using SSE transport, start a new conversation, and ask it to check the database status. The repository README lists each client's steps.
What licence is it under?
The repository does not state a licence. Confirm the terms with the maintainers before reusing or redistributing anything from it in your own products.
Is it production-ready?
The repository does not claim production readiness. It describes a hosted server open for public testing, so treat it as a starting point to review against your own security and availability needs.
Where does the data come from?
Provider data comes from the CMS NPPES registry, refreshed weekly according to the README, and exclusion data from the OIG exclusion list, refreshed on a monthly cycle. Always confirm exclusion hits at the official OIG site.
Using this in a governed AI setup?
Talk to us about running it inside the harness, with policy, guardrails and audit around it.