ramp-mcp is a Model Context Protocol server for retrieving and analyzing Ramp data or running tasks via the Ramp Developer API. To handle token and input size limitations, the server implements a simple ETL pipeline and an ephemeral SQLite database kept in memory for analysis by an LLM. By default, all requests target the demo environment; you can switch to production by setting RAMP_ENV=prd. The server exposes a suite of tools categorized as database tools (process_data, execute_query, clear_table), fetch tools (get_ramp_categories, get_currencies), and load tools that load various Ramp data into the ephemeral store. Each load tool maps to a specific Ramp scope (e.g., transactions:read, reimbursements:read, bills:read, locations:read, departments:read, bank_accounts:read, vendors:read, entities:read, limits:read, spend_programs:read, users:read) and enables data ingestion for subsequent queries and analysis. The README also notes performance considerations for large datasets and provides usage instructions for running via uv with Ramp client credentials, configuring environment variables, and optionally integrating with Claude Desktop. It emphasizes enabling the appropriate scopes on the Ramp client and including them when starting the server.