mcp-server-pp

MCP server for PharmaPendium data extraction — converts extraction results to XLSX, expands drug/concomitant rows, and uploads to Azure Blob Storage.

Tools

save_extraction

Converts a JSON array of extraction results into an XLSX file, uploads it to Azure Blob Storage, and returns a download URL (valid 7 days).

If the user provided corrections or new rules during the extraction session, pass them in notes — they are saved as a separate Notes sheet inside the XLSX for traceability.

Parameters:

• json_data (required): JSON array string — extraction results, one object per row
• file_name (optional): Output filename (with or without .xlsx). Defaults to extraction_results.xlsx
• notes (optional): User corrections, comments, or new rules applied in this session
• schema (optional): Output layout — auto (default), ae, or none

AE delivery layout. The LLM's JSON key order drifts between runs, so the column layout is fixed server-side. With schema: auto (the default) an AE frame is detected by its columns and rewritten to the PharmaPendium AE delivery headers, in this exact order:

FDAName, Subject, Sex, Age, Route, Dose, Dosage type, Adverse effect, Severity, File name, Page, Duration, Toxicity Parameter, Parameter Comment, Dose Comment, Metabolite name, Metabolite type, Comments, DataFormat

Extraction-schema names are renamed (Drug Name → FDAName, Species → Subject, Developmental Stage → Age, Dose/Concentration → Dose, Treatment Duration → Duration, Adverse Effect Comment → Parameter Comment, Substance Inducing AE → Metabolite name, Substance Inducing AE (Type) → Metabolite type); Comments and DataFormat are emitted empty. Already-normalized input passes through unchanged, unrecognized columns are appended after the schema columns (never dropped), and non-AE types (PK/EFF/CT/MET) are left alone. Pass schema: "none" to write the columns exactly as given.

expand_concomitants

Takes a JSON array of rows (from the built-in Excel tool) with Drug Name and Concomitants columns and rotates each drug into the title slot. The Concomitants cell is parsed as CT-style free text: drugs separated by ;, each entry shaped as Name (aliases), route, dose, dose_comment, duration. For each input row, emits N+1 rows — the promoted drug's parsed metadata populates Drug Name, Route, Dose/Concentration, Dose Comment, and Treatment Duration; the displaced drugs are re-serialized back into Concomitants. All other columns are copied unchanged. Semicolons inside (...) and semicolons followed by non-drug continuations (digits, or/and, bare units) are preserved within a single drug's metadata; bare-name segments at the start of a chunk are merged as combination-drug components (e.g. Elexacaftor; Tezacaftor; Ivacaftor (ELX; TEZ; IVA), orally, ... stays one entry).

Parameters:

• json_data (required): JSON array string of row objects — read via the built-in Excel tool
• file_name (optional): Output filename (with or without .xlsx). Defaults to expanded_results.xlsx

Example — one source row becomes three:

→

Required environment variable: AZURE_STORAGE_CONNECTION_STRING

Usage

With uvx (no installation needed)

uvx mcp-server-pp@1.3.0

MCP client config (e.g. Codemie, Claude Desktop)

{
  "mcpServers": {
    "pp-tools": {
      "command": "uvx",
      "args": ["mcp-server-pp@1.3.0"],
      "env": {
        "AZURE_STORAGE_CONNECTION_STRING": "${AZURE_STORAGE_CONNECTION_STRING}"
      }
    }
  }
}