The Upstage MCP Server provides a bridge between AI assistants and Upstage AI's document processing capabilities. It enables models to parse and extract information from a variety of document formats (PDFs, images, Office files, and more) while preserving structure and layout. Two primary tools are exposed: parse_document and extract_information. The parse_document tool takes a file_path and returns a processed representation of the document with preserved formatting. The extract_information tool accepts a file_path and an extraction schema (or auto-generates one by default) to return structured data extracted from the document. This makes it easier for downstream AI agents (like Claude) to locate and utilize specific data points such as invoice numbers, dates, or extracted table content. The server supports Claude Desktop integration, enabling seamless interaction via Claude's MCP interface.

Prerequisites:

• Python 3.10+ installed
• Node-based tooling not required for this Python-based server (uv/Astral UV is used for execution)
• Upstage API Key (for UPSTAGE_API_KEY)

Installation steps:

1. Clone the repository: git clone https://github.com/PritamPatil2603/upstage-mcp-server.git cd upstage-mcp-server

2. Install and set up uv (Astral UV) and the project in editable mode:

   Install uv if not already installed

   pip install uv

   Create and activate a virtual environment using uv

   uv venv

   Activate the virtual environment

   On Windows:

   .venv\Scripts\activate

   On macOS/Linux:

   source .venv/bin/activate

   Install dependencies in editable mode

   uv pip install -e .

3. Prepare environment variable (example):

   • Create a file or set environment variable UPSTAGE_API_KEY with your Upstage API key

4. Run the server (using uvx as recommended): uvx upstage-mcp-server

Notes:

• If uvx is not globally available, you may need to use the full path to uvx and adjust the mcp_config accordingly.
• For Claude Desktop integration, configure the claude_desktop_config.json as shown in the documentation to point to this MCP server.

Tips and caveats:

• Environment variables: Ensure UPSTAGE_API_KEY is set in the environment or provided in the MCP configuration file when using Claude Desktop.
• Output directories: Processed results are saved under ~/.upstage-mcp-server/outputs/document_parsing/ and ~/.upstage-mcp-server/outputs/information_extraction/, with schemas under the information_extraction/schemas directory.
• If you encounter a 'Server disconnected' error, verify uvx is installed and accessible in your PATH, or provide the absolute path to uvx in the configuration.
• Supported document formats include PDF, JPEG, PNG, TIFF, and common Office formats; ensure file size and page limits as per the server's expectations.
• When testing locally, you can use the two tools via the MCP interface:
  • parse_document: provide file_path to extract and preserve structure
  • extract_information: provide file_path and an optional schema_path or enable auto_generate_schema to true