Saksham

A lightweight MCP-style server that integrates with Google Docs and Gmail.

This project demonstrates how to build structured AI tool interfaces with approval gating, inspired by the Model Context Protocol (MCP).

---

✨ Features

• 📄 Append structured content to Google Docs
• 📧 Create Gmail drafts
• 🔐 OAuth-based Google authentication
• 🧠 MCP-style tool interface
• ✅ Human-in-the-loop approval before execution
• 🧩 Modular and extensible design

---

How to Use in Your Project

1. Run the MCP server: uvicorn server:app --reload

2. Call tools via API: POST /append_to_doc POST /create_email_draft

3. Integrate with your AI workflow:

   • Generate content using LLM
   • Send output to these endpoints

---

📁 Project Structure

google-mcp-server/
│── server.py
│── auth.py
│── docs_tool.py
│── gmail_tool.py
│── requirements.txt
│── credentials.json   (not committed)
│── token.json         (not committed)
│── README.md

---

⚙️ Setup

1. Clone the repository

git clone <your-repo-url>
cd google-mcp-server

2. Create virtual environment

python3 -m venv venv
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. 🔑 Google API Setup

1. Go to Google Cloud Console

2. Create a new project

3. Enable:

   • Google Docs API
   • Gmail API

4. Configure OAuth Consent Screen

5. Create OAuth Credentials (Desktop App)

6. Download credentials.json

7. Place it in project root

---

5. 🔐 Run OAuth

python3 auth.py

• Opens browser for login
• Generates token.json

6. ▶️ Run Server

uvicorn server:app --reload

Open:

http://127.0.0.1:8000/docs

🧪 How to Test

Append to Google Doc

Endpoint: POST /append_to_doc

Example:

{
  "doc_id": "YOUR_DOC_ID",
  "content": "Hello from MCP 🚀"
}

---

Create Email Draft

Endpoint: POST /create_email_draft

Example:

{
  "to": "[email protected]",
  "subject": "Test Draft",
  "body": "This is a test email"
}

🔄 Workflow

Request → Approval → Tool Execution → Response

• Every action requires manual approval
• Ensures safe and controlled execution

⚠️ Important Notes

Do NOT commit:

• credentials.json
• token.json

Approval is CLI-based (terminal input) Designed for local development

---

⚠️ Approval Flow

Every action requires manual approval in terminal:

ACTION: append_to_doc
PAYLOAD: {...}
Approve? (y/n):

Type:

y

---

🧠 MCP Design

This project demonstrates:

• Structured tool calls
• Separation of generation and execution
• Human approval before tool execution

---

❗ Notes

• No emails are sent automatically (draft only)
• Google Doc must have edit access
• Token is stored locally

---

📄 License

Apache License