Smart EHR MCP Server

# Core Server Logic This directory contains the core implementation of the SMART MCP server. ## Modules - `config.ts`: Handles loading and validation of server configuration (e.g., `config.json`). - `oauth.ts`: Implements the OAuth 2.0 Authorization Server logic, including endpoints for authorization, token exchange, client registration, and revocation. It defines the `MyOAuthServerProvider` which interacts with the MCP SDK. - `sessionUtils.js`: Manages user sessions, including creating sessions from EHR data, loading sessions from persisted storage (SQLite), and handling active session state in memory. - `sse.ts`: Sets up the main Express application, integrates the OAuth provider, defines API endpoints (like listing stored records), establishes the MCP Server-Sent Events (SSE) endpoint (`/mcp-sse`) for real-time communication, and registers the MCP tools. - `tools.js`: Defines the logic for the MCP tools (`grep_record`, `query_record`, `eval_record`) that operate on the patient's EHR data within an active session. ## Persistence Schema (SQLite) When EHR data is persisted using the utilities in `sessionUtils.js`, it is stored in an SQLite database specific to that session. ### FHIR Resources Table All FHIR resources are stored in a single table: ``` fhir_resources ``` With the following columns: - `resource_type` (TEXT): The FHIR resource type (e.g., "Patient", "Observation") - `resource_id` (TEXT): The resource ID - `json` (TEXT): The full JSON representation of the resource This approach allows for efficient querying by resource type and ID, while maintaining a simple database structure. ### Attachments Table A single table stores all attachments: ``` fhir_attachments ``` The attachments table has the following columns: - `id` (INTEGER): Auto-incrementing primary key - `resource_type` (TEXT): Type of FHIR resource the attachment belongs to - `resource_id` (TEXT): ID of the FHIR resource - `path` (TEXT): Path within the resource where the attachment was found - `content_type` (TEXT): MIME type of the attachment - `json` (TEXT): Original JSON representation of the attachment node - `content_raw` (BLOB): Raw binary data of the attachment (if available) - `content_plaintext` (TEXT): Extracted text content (if available) ## Example SQL Queries (for `query_record` tool) ### Find Patient Information ```sql SELECT resource_id, json FROM fhir_resources WHERE resource_type = 'Patient'; ``` ### Find Documents with "Diabetes" in Their Content ```sql SELECT a.content_plaintext, a.resource_id FROM fhir_attachments a WHERE a.resource_type = 'DocumentReference' AND a.content_plaintext LIKE '%diabetes%'; ``` ### Find Conditions with a Specific Code ```sql SELECT json FROM fhir_resources WHERE resource_type = 'Condition' AND json LIKE '%J45.909%'; ``` ### Join Patient and Observations ```sql SELECT p.json AS patient, o.json AS observation FROM fhir_resources p, fhir_resources o WHERE p.resource_type = 'Patient' AND o.resource_type = 'Observation' AND json_extract(o.json, '$.subject.reference') = 'Patient/' || p.resource_id; ``` ## Utility Functions The main functions for working with the database are: - `fetchAllEhrData(fhirBaseUrl, patientId, ehrAccessToken)`: Fetches all EHR data for a patient - `ehrToSqlite(fullEhr, db)`: Converts a FullEHR object to SQLite database tables - `sqliteToEhr(db)`: Reconstructs a FullEHR object from SQLite database ## Running the Server To start the MCP server, navigate to the project's root directory (where `package.json` is located) and run: ```bash bun run src/sse.ts ``` Or, to specify a different configuration file: ```bash bun run src/sse.ts --config /path/to/your/config.json ``` This will start the server using the configuration specified (by default `config.json` in the root directory). Make sure you have Bun installed and have run `bun install` to get the dependencies.