FastMCP Agent Orchestration

An MCP-powered agent orchestration project that solves a practical knowledge-work problem: information and actions are spread across Notion, Word documents, and the open web, but users need one interface to query, synthesize, and update content quickly.

Problem Solved

Teams frequently lose time context-switching between:

• internal notes in Notion,
• reports in Word documents,
• and external web references.

This project unifies those workflows using a tool-calling agent that can discover, read, summarize, and update content through MCP tools.

Solution Overview

The repository combines:

• a Streamlit chat client for interactive usage,
• a CLI client for terminal workflows,
• and an MCP server exposing Notion, Word, and web-search tools.

The LLM plans tool calls in sequence, executes them through MCP, and returns a concise final answer.

Architecture

┌──────────────────────┐
│ User (UI / CLI)      │
└──────────┬───────────┘
           │
           ▼
┌──────────────────────┐
│ Agent Client         │
│ - app.py (Streamlit) │
│ - client.py (CLI)    │
└──────────┬───────────┘
           │ MCP streamable-http
           ▼
┌──────────────────────┐
│ MCP Server           │
│ - server.py          │
│ Tools: Notion, Word, │
│ Tavily Web Search    │
└──────────────────────┘

Repository Structure

FastMCP---Agent-Orchestration/
├── app.py                          # Streamlit-based agent chat UI
├── client.py                       # CLI interactive agent client
├── server.py                       # FastMCP server with Notion/Word/Web tools
├── requirements.txt
└── Office-Word-MCP-Server/         # Word MCP package and docs

Key Features

• Autonomous multi-step tool calling via MCP
• Notion discovery and content reading tools
• Notion content append/update tools
• Word document create, read, append operations
• Tavily-based web search integration
• Streamlit interface with tool trace visibility
• CLI mode for quick local testing

Prerequisites

• Python 3.10+
• Access to:
  • Azure/OpenAI-compatible chat endpoint
  • Notion integration key and API access
  • Tavily API key (for web search tool)

Setup

1) Clone and install dependencies

git clone https://github.com/Rahul-Ganesan/FastMCP---Agent-Orchestration.git
cd FastMCP---Agent-Orchestration
pip install -r requirements.txt

2) Create .env

OPENAI_API_KEY=your_key
AZURE_ENDPOINT=your_azure_openai_endpoint
AZURE_DEPLOYMENT_NAME=your_chat_deployment_name

NOTION_API_KEY=your_notion_integration_key
NOTION_BASE_URL=https://api.notion.com/v1
NOTION_VERSION=2022-06-28

TAVILY_API_KEY=your_tavily_key

# Optional override (default shown)
MCP_SERVER_URL=http://127.0.0.1:8050/mcp

Run

Terminal 1: Start MCP server

python server.py

Terminal 2: Start Streamlit app

streamlit run app.py

Then open the Streamlit local URL in your browser.

Optional: Run CLI client instead of Streamlit

python client.py

Example Use Cases

• "Search Notion pages for Q1 planning notes and summarize key action items."
• "Read page content for the design retrospective and extract risks."
• "Create a Word doc named weekly_update and add a concise status summary."
• "Use internal notes plus web context to suggest implementation options."

Troubleshooting

• If the UI cannot call tools, verify server.py is running and MCP_SERVER_URL is correct.
• If model calls fail, verify OPENAI_API_KEY, AZURE_ENDPOINT, and AZURE_DEPLOYMENT_NAME.
• If Notion calls fail, confirm integration access and NOTION_API_KEY.
• If web search fails, check TAVILY_API_KEY.

Notes

• Office-Word-MCP-Server/README.md documents advanced Word-server capabilities and packaging details.
• This root README focuses on orchestration flow for this specific repo.