Health plan data
you can cite.
LLMs can extract benefit data. Only Health Plan API guarantees every value is bound to its source line by an independent layout engine, never the extracting model's self-report. Built on the open Benefit Plan Standard.
The Coverage Question
"Is this drug covered?" Answered, with a citation.
For a given plan and drug: is it covered, under which benefit, at what tier, with which restrictions? Every value in the answer links back to the exact line of the source document. That is the unit of value we sell, and no raw LLM can produce it about itself.
Formulary-aware
Formulary tiers, retail and mail-order networks, prior authorization, step therapy, quantity limits, and indication-dependent coverage are modeled as structured data, not paragraphs. GLP-1s and other high-cost drugs are exactly where this precision pays for itself.
Cited line by line
When one table cell carries three prices, each price resolves to its own line in the source PDF. Your support team, your compliance team, and your AI agent all see the same evidence trail.
Built for the front line
Provider offices, digital health platforms, and pharmacy teams burn hours on manual coverage lookups. A cited answer in seconds replaces a phone call, a PDF hunt, and a guess.
Three Surfaces, One Engine
The same cited answer, wherever you need it.
One resolver produces the answer. Three surfaces deliver it, all carrying the same field-level provenance.
Portal, for people
Ops and admin teams look up a plan, see the cited answer, and click any value to open the source PDF at the exact highlighted line. No training, no query language.
API, for systems
A REST call returns the full BPS-conformant plan document, with a provenance sidecar that maps every output field to its source location. Built for enrollment platforms, quoting engines, and data pipelines.
MCP, for AI
Any MCP-compatible agent gets the same cited answers as first-class tools. An AI that answers coverage questions without provenance is a liability. One connected to Health Plan API cites its sources.
The Platform
Two things you don't get anywhere else.
The extraction layer is commoditizing. Citation integrity and an open standard are not, and they compound.
Independent citation layer
Every copay, limit, and exclusion points to a page and bounding box produced by an independent layout engine, not the model's self-report. The difference matters when a regulator or attorney asks where a value came from.
Open standard, not a walled garden
We're the reference implementation of the Benefit Plan Standard, the vendor-neutral schema for health plan content. Your data model isn't locked to our roadmap.
Extraction pluggable, citations permanent
We'll use the best extractor on the market as models evolve. What doesn't change is where citations come from. The provenance layer is architectural, not vendor-bound.
AI-agent ready
Connect any MCP-compatible AI agent directly to benefit plan data. Query plans, compare benefits, validate data through natural language. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible tool.
Why It Matters
The deposition test.
When a regulator or plaintiff's attorney asks "where did this value come from?", what does your vendor show? A screenshot annotation from the same model that generated the value, or an independent layout trace? Ours is the second.
Regulation-ready
California SB 942 and Texas HB 149 AI provenance laws are in force, with Connecticut and Colorado following in 2026 and 2027. Digital provenance is becoming the enterprise baseline. Our architecture was designed for this floor from day one.
Not UI-only
Hover-to-source UI is easy. An independent citation layer that survives a model swap, a pipeline rewrite, and an audit is not. Most vendors ship the first. We ship the second.
Every field, every stage
Page number, bounding box, source text, extraction stage, and confidence score are persisted with every value. Your UI, your audit log, your customer disputes: all trace to the same record.
Append-only by design
Traceability records are never edited in place. Corrections create a new version with its own provenance chain. Plans are immutable; the audit trail is complete.
MCP Integration
Built for AI-assisted development.
Developers using AI coding assistants can connect directly to Health Plan API through our MCP Server. Query benefit data in plain language, right from your editor.
Connect in minutes
Add the MCP Server to your Claude Code, Cursor, or Windsurf config and your AI agent gets direct access to normalized plan data.
{
"mcpServers": {
"healthplanapi": {
"command": "npx",
"args": ["-y", "@healthplanapi/mcp-server"],
"env": {
"HEALTHPLANAPI_KEY": "your_api_key"
}
}
}
}Available MCP tools
The server exposes structured tools that let agents query and reason over benefit data without writing raw HTTP calls.
search_plans: Filter by carrier, year, or benefit criteria.get_plan_summary: Concise plan overview with key cost shares and validation status.get_plan_benefits: Full benefit tree for a plan ID.get_copay: Copay and coinsurance for a specific benefit.get_formulary: Pharmacy and drug-tier coverage for a plan.compare_plans: Side-by-side comparison of two to five plans.validate_plan: Validate a plan against the Benefit Plan Standard schema.list_carriers: Available carriers and plan counts.
Natural language queries
Ask your AI assistant: "What is the in-network deductible for this PPO?" and get a structured answer pulled directly from validated plan data, with the source citation attached.
No custom client needed
MCP handles the transport layer. Your agent calls tools, not endpoints. No SDK setup, no request serialization, no response parsing.
Same auth, same data
Your existing API key works with the MCP Server. The same tenant isolation, rate limits, and data access rules apply as with the REST API.
Build on data you can cite.
We are onboarding a small number of design partners: organizations with real, recurring coverage-lookup pain who want cited answers before anyone else. Provider groups, digital health platforms, and pharmacy teams are a strong fit.