Google MCP Agent

🚀 Google MCP Agent

A unified FastMCP-powered agent that connects Gmail, Google Drive, and Google Classroom with LLM-ready tools for automation, parsing, and intelligent retrieval.

---

📌 What This Repo Offers

• ✅ Read and manage Gmail (inbox, unread, spam)
• ✅ Read file content from Google Drive (Docs, PDFs, TXT)
• ✅ List Google Classroom courses
• ✅ LLM-compatible tools exposed via FastMCP
• ✅ Easily deployable with uv in an agentic RAG pipeline

---

⚙️ Setup Guide

Before running the project, set up your Google Cloud project for OAuth.

🔑 1. Enable APIs in Google Cloud Console

1. Go to GCP Console
2. Navigate to "APIs & Services" > "OAuth consent screen"
3. Choose "External" > Fill in app name and support email
4. Under Scopes, you don't need to add anything manually now
5. Under Test users, add the Gmail/Drive account you want to access (must match the email you're using)

🧾 2. Create OAuth 2.0 Credentials

1. Go to "APIs & Services" > "Credentials"
2. Click "Create Credentials" > "OAuth client ID"
3. Choose "Desktop App" or "Web App"
4. Download the client_creds.json file

🔐 Authentication Info (One-time Manual, Then Persistent)

• The first time you run the server, a browser window will open asking you to log in and authorize access.
• After successful login, a token.json file will be created automatically in your local directory.
• This file contains your OAuth access/refresh token and is reused for future runs.
• You won’t need to re-authenticate unless the token expires or is deleted.

⚠️ _No need to publish the app — keeping it in testing mode is fine as long as the user is whitelisted._

---

🛠️ Configuration Example

Add this block to your msp.json or equivalent MCP config to run the main server:

{
    "mcpServers": {
      "main_mcp_server": {
        "command": "uv",
        "args": [
          "--directory",
          "absolute_path\\to\\current_project",
          "run",
          "server.py",
          "--creds-file-path",
          "absolute_path\\to\\client_creds.json",
          "--token-path",
          "absolute_path\\to\\app_tokens.json"
        ]
      }
}

🧰 Available Tools (Exposed via FastMCP) These tools are registered with FastMCP and can be used directly by your copilot system or any agent pipeline.

📬 Gmail Tools 🔹 get_unread_emails(limit: int = 5) Returns a list of unread emails from the user's inbox with metadata like subject, sender, and snippet. Automatically marks them as read.

🔹 read_email(email_id: str) Reads and decodes the full plain-text body of a specific email by its ID.

🔹 get_read_emails(limit: int = 50) Fetches recently read (non-unread) emails from the inbox.

🔹 get_spam_emails(limit: int = 20) Retrieves a list of emails from the Gmail spam folder with minimal metadata.

📂 Google Drive Tools

🔹 list_my_drive_files(limit: int = 10) Returns a list of recent files in the user's Google Drive, including ID, name, MIME type, and last modified time.

🔹 read_file_content(file_id: str) Reads and extracts text from a file in Drive. Supports:

Google Docs (exported as plain text)

Plain .txt files

PDFs (parsed using PyPDF2)

🏫 Google Classroom Tool 🔹 list_courses() Fetches the list of enrolled Google Classroom courses (read-only).

Each tool can be invoked via MCP runtime or LLM-powered interfaces. This setup enables rich, context-aware workflows that interact with the user's real data.