PDF RAG MCP Server

# MCP CLI A command-line interface for interacting with Model Context Protocol (MCP) servers. ## Features - **Tool Discovery**: List and describe available tools from MCP servers - **Tool Invocation**: Call tools with parameters from the command line - **Flexible Configuration**: JSON config files with CLI overrides - **Parameter Support**: Simple values, JSON files, and file references - **Output Formats**: Human-readable and JSON output modes ## Installation ```bash pip install -r requirements.txt ``` ## Configuration Create a `server-config.json` file in your working directory: ```json { "server_path": "/path/to/mcp_server.py", "server_args": ["--db-path", "/path/to/database"], "timeout": 30 } ``` You can also specify a different config file with `--server`: ```bash python mcp_cli.py --server /path/to/config.json list-tools ``` ## Usage ### List Available Tools ```bash python mcp_cli.py list-tools ``` With JSON output: ```bash python mcp_cli.py --json list-tools ``` ### Describe a Tool ```bash python mcp_cli.py describe <tool-name> ``` Example: ```bash python mcp_cli.py describe pdf_add ``` ### Call a Tool ```bash python mcp_cli.py call <tool-name> [param=value ...] ``` Examples: ```bash # Simple parameters python mcp_cli.py call pdf_search_similarity query="machine learning" top_k=5 # File reference (loads file content) python mcp_cli.py call pdf_add pdf_path=@/path/to/document.pdf # JSON file parameter python mcp_cli.py call some_tool config=@config.json # JSON output python mcp_cli.py --json call pdf_list response_format=json ``` ### Parameter Syntax - **Simple values**: `key=value` (strings, numbers, booleans) - Booleans: `active=true` or `active=false` - Numbers: `age=30` or `price=19.99` - Strings: `name=John` - **File references**: `key=@filename` - JSON files are automatically parsed - Other files are read as text/binary content - Example: `document=@paper.pdf` ## Examples with PDF RAG Server ```bash # List all tools python mcp_cli.py list-tools # Add a PDF to the database python mcp_cli.py call pdf_add pdf_path=@/path/to/paper.pdf chunk_size=3 overlap=1 # Search for content python mcp_cli.py call pdf_search_similarity query="neural networks" top_k=5 response_format=markdown # List all PDFs in database python mcp_cli.py call pdf_list response_format=markdown # Get JSON output for scripting python mcp_cli.py --json call pdf_list response_format=json ``` ## Exit Codes - `0`: Success - `1`: Error (see stderr for details) ## Design Decisions - **Stateless**: Each invocation connects fresh to the server - **Default Config**: Uses `./server-config.json` by default - **Explicit Parameters**: File references require `@` prefix - **Type Inference**: Automatically parses booleans and numbers - **Error Handling**: Clear error messages to stderr with proper exit codes