Overview
What is MCP APP?
MCP APP is an MCP server application that provides RAG (Retrieval-Augmented Generation) and web searching tools for LLMs. It allows the LLM to retrieve knowledge from a vectorstore and add documents, augmenting the knowledge available to the model. It is designed for use with MCP clients such as Claude Desktop.
How to use MCP APP?
For development, run uv sync, then source .venv/bin/activate, followed by mcp dev run. To integrate with Claude Desktop, run mcp install server.py --env-file .env --with sqlalchemy --with pgvector --with openai --with "psycopg[binary]" --with pydantic --with python-dotenv --with tavily-python.
Key features of MCP APP
- RAG tools for knowledge retrieval and document addition.
- Web searching tools for real-time information access.
- Uses PostgreSQL with PGVector as vectorstore.
- OpenAI embeddings for vector representations.
- Integrates with Claude Desktop via MCP protocol.
Use cases of MCP APP
- Enabling LLMs to retrieve and add documents to a vectorstore.
- Performing web searches to supplement LLM knowledge.
- Using Claude Desktop to interact with custom RAG pipelines.
- Augmenting existing knowledge bases with new documents on the fly.
FAQ from MCP APP
How do I run MCP APP in development?
Use uv sync, then source .venv/bin/activate, followed by mcp dev run.
How do I install MCP APP for Claude Desktop?
Run the command: mcp install server.py --env-file .env --with sqlalchemy --with pgvector --with openai --with "psycopg[binary]" --with pydantic --with python-dotenv --with tavily-python.
What dependencies does MCP APP require?
The README lists SQLAlchemy, pgvector, OpenAI, PostgreSQL, and additional packages: psycopg (binary), pydantic, python-dotenv, and tavily-python.
Does MCP APP need a database?
Yes, it uses PostgreSQL as the SQL database and PGVector as the vectorstore, with SQLAlchemy for ORM interaction.
