Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Claims desk MCP Server What are the Medicare documentation requirements for CPT 99213?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Personal demo by Andrew Espira.
A claims reviewer MCP App: one tool returns structured prior-auth / attachment / filing guidance plus doc:// citation URIs (never document bytes). The App streams the live CMS LCD or claims-manual pages beside that answer.
Implements the current MCP Apps extension (io.modelcontextprotocol/ui).
Check | Where |
Tool bound to a UI |
|
UI resource |
|
Capability |
|
App-only tools |
|
Split | Tool JSON has citation URIs only; the App fetches bytes via app-only tools |
Related MCP server: Medical Billing MCP
Python 3.10+ (3.12 recommended). Needs network for live cms.gov fetches. PDF page preview uses pdftoppm (poppler).
python3.12 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e .
python scripts/serve_app.pyLeave Medicare / 27447, click Ask MCP tool.
Confirm First Coast Jurisdiction N + 12-month / 1 calendar year filing on the left, LCD L33618 on the right.
Try 70553 (claims-manual PDF), E0601 (no DMEPOS PA; CMNs discontinued), and UHC commercial outpatient MRI (UHC PA list, not a Medicare manual).
The local host resources/reads the ui:// HTML, loads it in an iframe, and proxies tools/call over postMessage. The view never POSTs /mcp.
python -m mcp_pdf_demo.serverPoint the client at that command. Optional Streamable HTTP: python -m mcp_pdf_demo.server --http → http://127.0.0.1:3001/mcp.
Optional enrichment (not required to review the Apps wiring):
export RCI_API_KEY=...
export RCI_API_URL=https://api-dev.rcintell.comAlways two parts: a tool that returns data, and a ui:// resource the host renders.
Python SDK Apps extension — Apps() + MCPServer(..., extensions=[apps]) so the server advertises io.modelcontextprotocol/ui.
Bind the result tool — @apps.tool(resource_uri="ui://claims-desk/app.html", meta={"ui/resourceUri": "..."}) on review_claim_requirements.
Serve the view — FunctionResource / add_html_resource at that URI with MIME text/html;profile=mcp-app. Bundle CSS + JS into one HTML file (app_html.py + app/view.js).
App-only helpers — same resource_uri, visibility=["app"], for chunked reads and PDF rasterization.
View talks to the host, not the server — ui/initialize then tools/call over window.parent.postMessage (app/view.js).
Host is the MCP client — fetch ui:// HTML, iframe it, handle ui/initialize, proxy tools/call / resources/read (app/host.js + scripts/serve_app.py). Production hosts (Claude, VS Code, ext-apps basic-host) do this step for you.
Policy URLs live in fixtures/catalog.json. Fetch/cache is src/mcp_pdf_demo/live.py.
Path | Role |
| MCP server: Apps tools, |
| Live CMS fetch + optional RCI |
| Bundles the view into one HTML document |
| MCP App view ( |
| Local Apps host |
| Runs that host on |
| CPT → CMS URL map |
MIT
List live CMS policy documents wired into this claims desk.
Inputs: no named parameters. Returns: result
App-only: rasterize the first pages of a cited policy PDF for the reviewer.
Inputs: uri (required), pages. Returns: result
App-only: stream a fetched policy document in base64 chunks (offset/length/hasMore). Used by the claims App to render source docs.
Inputs: uri (required), length, offset. Returns: result
Resolve claim documentation requirements for a payer + CPT. Fetches live cataloged policy sources (CMS or payer) and optionally enriches from an RCI knowledge API when RCI_API_KEY is set. Returns structured requirements plus document URIs — never PDF/HTML bytes.
Inputs: cpt (required), payer (required). Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 4 tools
Each tool has a distinct primary purpose, and the names/descriptions are mostly clear. However, read_document_bytes and preview_pdf_pages both operate on cited policy documents for app rendering, and their raw-bytes vs. rasterized-preview distinction could cause minor selection confusion.
All tools follow a consistent verb_noun pattern: review_claim_requirements, read_document_bytes, preview_pdf_pages, and list_policy_library. No casing or style inconsistencies exist, and each name clearly suggests its action.
Four tools is well-scoped for a focused claims document and requirements server. Each tool covers a distinct step in the workflow: resolving requirements, reading document bytes, previewing PDFs, and listing the policy library.
The core workflow is covered end-to-end: review claim requirements, access document content, preview PDFs, and browse the policy library. Minor gaps remain, such as no general search/filter over the library and no non-app-facing document retrieval path.
Deterministic CMS NCCI/MUE claim scrubbing and CARC/RARC denial explanations, cited to source.
Medicare national and local coverage policy from the official CMS Coverage API.
Medicare coverage and fee-schedule amounts for CPT/HCPCS codes against a diagnosis
DME billing and medical coding: code lookup, DMEPOS fee schedules, Medicare prior-auth checks