MedPrice AI

mcp-medprice-ai

A hosted MCP server exposing US hospital chargemaster cost data to AI assistants.

• MCP endpoint: https://mcp.medprice.ai/mcp
• gRPC backend: api.medprice.ai:443

Install in Claude Code

claude mcp add --transport http mcp-medprice-ai https://mcp.medprice.ai/mcp

To share with everyone in a project, add --scope project (writes to .mcp.json). To make it available across all your projects, use --scope user.

Usage

The server exposes five tools:

• list_hospitals — returns the supported hospitals with their hospital_id, EIN, name, structured_locations (addresses with geocoded coordinates where available), last_updated_on, and revision history (each revision's date, revision_id, and whether it has payer-specific rate data).
• get_hospital_chargemaster_cost — looks up cost stats for a billing code at a single hospital.
• list_hospital_code_costs — looks up cost stats for a billing code across every hospital that has a matching chargemaster entry, paginated. Use this instead of calling get_hospital_chargemaster_cost once per hospital when comparing prices for the same procedure across hospitals.
• get_code_price_distribution — returns the distribution of every hospital's price for a billing code (percentiles, histogram, and the real hospital at each percentile), computed server-side. Use instead of paging through list_hospital_code_costs for "is this price high?" questions.
• list_code_types — lists every distinct billing code type (e.g. CPT, MS-DRG) with catalogued cost data, along with each type's distinct code count and total hospital reports.
• list_codes — lists every distinct code under a given code type, paginated, with a raw chargemaster description and reporting-hospital count per code. Use this to discover which codes exist under a code system before pricing them.

The typical flow is to call list_hospitals first to discover available hospitals and their IDs, then call get_hospital_chargemaster_cost with the desired hospital_id — or call list_hospital_code_costs directly when the question is about a code across hospitals rather than one specific hospital. Once installed, you can just ask your assistant something like:

What's the fee schedule cost of MS-DRG 652 at Medical City Alliance?

The assistant will call list_hospitals to find the hospital's ID, then call get_hospital_chargemaster_cost with:

{  "hospital_id": "3f9a1c2b8e4d5f6a7b8c9d0e1f2a3b4c",  "code_type": "MS-DRG",  "code": "652",  "methodologies": ["fee schedule"]}

and returns:

{  "hospital": "MEDICAL CITY ALLIANCE",  "found": true,  "hospital_id": "3f9a1c2b8e4d5f6a7b8c9d0e1f2a3b4c",  "results": [    {      "methodology": "fee schedule",      "found": true,      "cost": {        "code_type": "MS-DRG",        "code": "652",        "min": "26851.11",        "max": "190885.00",        "avg": "34387.70",        "median": "28084.07",        "std_dev": "13735.94"      },      "description": {        "hospital_name": "MEDICAL CITY ALLIANCE",        "location": "3101 N Tarrant Pkwy, Fort Worth, TX, 76177",        "code_description": "KIDNEY TRANSPLANT",        "methodology_note": "fee schedule"      }    }  ]}

Tool reference

list_hospitals

• page_size (optional) — maximum number of hospitals to return. Defaults to 500 (the entire current registry in one call), capped at 500.
• page_token (optional) — opaque token from a previous response's next_page_token, for pagination. If next_page_token is non-empty, keep calling with it until it's empty rather than assuming one page is the full list.

Each hospital's revisions array now includes a revision_id per revision (in addition to revision_date and has_payer_data) — pass it as get_hospital_chargemaster_cost's revision_id to price that specific past revision instead of the hospital's latest one.

get_hospital_chargemaster_cost

• hospital_id (required) — opaque hospital identifier from list_hospitals.
• code_type (required) — code system, e.g. APR-DRG, CDM, CPT, HCPCS, MS-DRG, RC. Hospitals may also support additional proprietary code types.
• code (required) — the billing/chargemaster code.
• methodologies (optional) — array of case rate, fee schedule, other, percent of total billed charges, per diem. One result per entry, in request order, each with its own found/cost/description. Omit or pass an empty array to get a single aggregate result across all methodologies.
• revision_id (optional) — a revision_id from list_hospitals, to price that specific past revision instead of the hospital's latest one.

Returns hospital, found (true if at least one entry in results was found), hospital_id, and results — one entry per requested methodology (or one aggregate entry when methodologies is omitted), each shaped like { methodology, found, cost, description }.

list_hospital_code_costs

Like get_hospital_chargemaster_cost, but returns one result per hospital that has a matching chargemaster entry for the code, instead of requiring a hospital_id up front — useful for "which hospital is cheapest for X" questions without a list_hospitals + N × get_hospital_chargemaster_cost round trip.

• code_type (required) — same as above.
• code (required) — same as above.
• methodology (optional) — single methodology string, same options as above. Unlike get_hospital_chargemaster_cost, this filter stays singular — it still only aggregates or narrows to one methodology, it doesn't fan out to multiple results entries per hospital.
• page_size (optional) — maximum number of results to return. Defaults to 500 (every matching hospital in one call at the current registry size), capped at 500.
• page_token (optional) — opaque token from a previous response's next_page_token, for pagination. If next_page_token is non-empty, keep calling with it until it's empty rather than assuming one page is the full list.

Returns results (each hospital shaped like a get_hospital_chargemaster_cost response — hospital, found, hospital_id, and a results array with exactly one entry given the singular methodology filter) and next_page_token. Only hospitals with a matching entry (their latest revision) are included — there are no found: false entries.

get_code_price_distribution

Distribution of one per-hospital price figure across every hospital reporting the code, computed server-side in one call.

• code_type (required), code (required) — same as above.
• methodology (optional) — same options as list_hospital_code_costs; omit to aggregate across all.
• min_sample_count (optional) — same semantics as list_hospital_code_costs.
• bin_count (optional) — histogram bars; defaults to 16, max 200.
• reference_price (optional) — when positive, hospitals_below_reference counts hospitals priced strictly below it.
• statistic (optional) — PRICE_STATISTIC_AVERAGE (default), PRICE_STATISTIC_MEDIAN, PRICE_STATISTIC_MIN or PRICE_STATISTIC_MAX: the per-hospital figure being distributed.

Returns hospital_count, min/p10/q1/median/q3/p90/max, std_dev, histogram (histogram_lo, histogram_hi, histogram_bins, overflow_count), a *_hospital point (hospital_id, hospital_name, location, price) for each percentile, and hospitals_below_reference. hospital_count 0 means no hospital reports a price.

list_code_types

No arguments. Returns code_types, one entry per distinct code type present in the catalog (most code-rich first), each with code_type, code_count (distinct codes catalogued under that type), and total_hospital_reports (sum of hospital_count across every code under that type — not a distinct-hospital count). Unpaginated.

Use this to discover which code systems have data before drilling into list_codes.

list_codes

• code_type (required) — code system to list codes for, e.g. CPT. From list_code_types.
• page_size (optional) — maximum number of results to return. Defaults to 500, capped at 500.
• page_token (optional) — opaque token from a previous response's next_page_token, for pagination. If next_page_token is non-empty, keep calling with it until it's empty rather than assuming one page is the full list.

Returns codes (each with code, raw_description — raw chargemaster text, not necessarily a human-readable procedure name — and hospital_count, distinct hospitals reporting that code on their latest revision), next_page_token, and total_count (total matching codes across all pages). Use this to discover which codes exist under a code system before pricing them with get_hospital_chargemaster_cost or list_hospital_code_costs.

Development

Run locally against the production gRPC backend

GRPC_HOST=api.medprice.ai:443 npx tsx src/index.ts

Run locally against a custom gRPC backend

GRPC_HOST=<host:port> npx tsx src/index.ts

Run as HTTP server

TRANSPORT=http GRPC_HOST=api.medprice.ai:443 npx tsx src/index.ts

All MCP requests go to POST /mcp. PORT defaults to 3000.

Test gRPC connectivity

GRPC_HOST=api.medprice.ai:443 npx tsx src/test.ts

Docker

Build

docker build -t mcp-medprice-ai .

Run

docker run --rm -p 3000:3000 \  -e GRPC_HOST=api.medprice.ai:443 \  mcp-medprice-ai

TRANSPORT=http and PORT=3000 are set by default in the image. Override PORT if needed:

docker run --rm -p 8080:8080 \  -e GRPC_HOST=api.medprice.ai:443 \  -e PORT=8080 \  mcp-medprice-ai