Custom MCP Server with RAG & Tools

This project implements a Custom MCP (Multi-tool Control Protocol) server using FastMCP, supporting:

• Notes Management (Add / Read / Summarize Notes)
• Weather Search (via WeatherAPI)
• Brave News Search (via Brave API)
• Document Ingestion & RAG Queries (via GroundX)
• OpenAI GPT Integration for Summarization & RAG

---

Features

• Append, read, and summarize text notes locally
• Get real-time weather info
• Search current news headlines via Brave API
• Ingest PDFs and perform RAG-based semantic search using GroundX
• Supports OpenAI GPT (e.g., gpt-4o) for completions
• Local MCP Inspector for testing/debugging

---

Installation

1. Using uv (Recommended)

uv lock
uv sync

2. Using pip

pip install -r requirements.txt

---

Environment Variables (.env)

OPENAI_API_KEY=your_openai_api_key
GROUNDX_API_KEY=your_groundx_api_key
WEATHER_API_KEY=your_weatherapi_key
BRAVE_API_KEY=your_brave_api_key
BUCKET_ID=your_groundx_bucket_id

---

Running the MCP Server

Development Mode (with Inspector UI)

mcp dev main.py

If you get authentication errors, run with:

DANGEROUSLY_OMIT_AUTH=true mcp dev main.py

Development Mode with claude desktop app:

mcp install main.py
mcp run main

---

Available Tools & Resources

Tool Name	Description
add_note(message)	Append note to local file
read_notes()	Return all stored notes
note_summary_prompt()	Generate a prompt to summarize notes using GPT
brave_search_results(q)	Latest news via Brave Search API
fetch_weather(city)	Real-time weather from WeatherAPI
ingest_documents(path)	Upload PDF to GroundX knowledge base
process_search_query(q)	Perform RAG search with OpenAI GPT completions
search_doc_for_rag_context(q)	Retrieve context text for GPT queries

---

API Example (via Inspector)

{
  "tool": "fetch_weather",
  "args": {
    "city": "New York"
  }
}

---

Example Workflow

uv sync
cp .env.example .env  # setup .env file
mcp dev main.py       # start MCP development server

---

Troubleshooting

Error: Connection Error - Missing Proxy Token

➡️ Solution: Use the URL provided or run:

DANGEROUSLY_OMIT_AUTH=true mcp dev main.py

License

MIT License © 2025 Your Name / Your Organization

---

Generated with ❤️ using MCP, GroundX, OpenAI.