Skip to content

README.md

Latest commit

 

History

History
35 lines (24 loc) · 1.06 KB

README.md

File metadata and controls

35 lines (24 loc) · 1.06 KB
Raw

Items MCP Server

Model Context Protocol (MCP) server exposing product retrieval as tools for AI agents.

Purpose

MCP standardizes how agents discover and invoke tools across services. This server runs independently (port 8001 on host) so agents can call get_formatted_items_context via HTTP. Used by notebooks/week4/04-MCP.ipynb.

Architecture

items_mcp_server/
├── src/items_mcp_server/
│   ├── main.py      # FastMCP app, @mcp.tool() get_formatted_items_context
│   ├── utils.py     # retrieve_items_data, process_items_context (Qdrant hybrid search)
│   └── core/config.py
├── Dockerfile
└── pyproject.toml

Tool

  • get_formatted_items_context(query, top_k=5): Returns formatted product descriptions from Amazon-items-collection-01-hybrid-search. Same logic as api.agents.tools.get_formatted_items_context.

Running

# Via Docker Compose (with api, streamlit, qdrant)
docker compose up items_mcp_server

# Standalone
uv run python -m items_mcp_server.main

Port: 8001 (host) → 8000 (container)