Coffee MCP Server

What is Coffee MCP Server?

Coffee MCP Server is a robust FastAPI-based API server for asynchronous document extraction and processing. It extracts text and tables from PDFs and images using OCR, generates text embeddings via OpenAI or Anthropic, and stores job results in MongoDB for persistent tracking. It is designed for developers who need to automate document data extraction at scale.

How to use Coffee MCP Server?

The server requires Python 3.8+, a running MongoDB instance, Tesseract OCR, and Poppler. After cloning the repository, install system and Python dependencies, configure environment variables in a .env file (including MONGODB_URI, OPENAI_API_KEY, and optionally ANTHROPIC_API_KEY), and start the server with uvicorn app:app --host localhost --port 8000 --reload. Documents are submitted via the POST /v1/extract_data endpoint; the returned job_id is used to poll GET /v1/extract_data_job for status and retrieve results from GET /v1/extract_data_result.

Key features of Coffee MCP Server

• Asynchronous document processing with non‑blocking API
• Page‑by‑page PDF processing for real‑time status updates
• OCR‑based text extraction using Tesseract
• Table detection and extraction
• Text embedding generation via OpenAI or Anthropic
• MongoDB storage for persistent job tracking and results

Use cases of Coffee MCP Server

• Digitize and extract content from scanned PDFs or images
• Automate data extraction from invoices, forms, or reports
• Build retrieval‑augmented generation pipelines with embedded document chunks
• Process large document sets (500+ pages) with progress tracking
• Monitor long‑running extraction jobs with polling and paginated results

FAQ from Coffee MCP Server

What kind of document processing does Coffee MCP Server perform?

It extracts text using OCR, detects and extracts tables, and generates text embeddings. Processing runs asynchronously on a background thread, enabling page‑by‑page updates for large PDFs.

What are the required dependencies and runtime environment?

You need Python 3.8+, a running MongoDB instance, Tesseract OCR engine, and Poppler. API keys for OpenAI (required) and optionally Anthropic are set in a .env file.

Where are processed documents and results stored?

All job metadata, progress, extracted text, tables, and embeddings are stored in MongoDB. The database and collection are configured via the MONGODB_URI and MONGODB_DATABASE environment variables.

What transports and authentication does the server use?

The server exposes a RESTful HTTP API via FastAPI/Uvicorn. No built‑in authentication is described; API keys for embedding providers are stored locally in the environment file. CORS is configured on the server.

Are there any known limits or performance considerations?

Processing very large documents (500+ pages) requires sufficient memory for OCR operations. The background thread architecture keeps the API responsive, and debug images can be disabled by omitting the RAGNOR_DEBUG_IMAGES_PATH variable.