chatgpt-custom-mcp-for-local-files

adamgivon/chatgpt-custom-mcp-for-local-files

3.2

If you are the rightful owner of chatgpt-custom-mcp-for-local-files and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

This project provides a custom Model Context Protocol (MCP) server for reading local files securely using Cloudflare Tunnel and OAuth 2.0 authentication.

Tools
4
Resources
0
Prompts
0

ChatGPT Custom MCP for Local Files

Custom MCP server for reading local files using Cloudflare Tunnel. Point ChatGPT (chatgpt.com, not api) to a dedicated folder on your machine - the model reads complete files on demand. No RAG preprocessing, no partial reads, no uploads. Direct file access through secure OAuth 2.0.

To be clear, file contents are transmitted to OpenAI servers for processing during conversations. They are not permanently stored in ChatGPT Projects, but are subject to OpenAI's standard conversation data retention policies. This approach differs from manual uploads in that files are fetched on-demand rather than pre-uploaded.

Features

  • Secure OAuth 2.0 authentication with dynamic client registration (RFC 7591)
  • Cloudflare Tunnel for secure remote access without port forwarding
  • Complete file access - ChatGPT reads entire files, not chunks
  • Smart file discovery - Search by name, list directories, read on demand
  • Persistent sessions - Registered clients survive server restarts
  • systemd integration - Start/stop with system services
  • Desktop launchers - GUI shortcuts for easy control (Linux)

Why This Over RAG?

Traditional RAG (Retrieval-Augmented Generation):

Requires preprocessing and embedding generation. Returns partial/chunked content. Adds latency and complexity. Limited context about file structure.

This MCP Server:

Direct file system access. Reads complete files on demand. ChatGPT can explore your directory structure. Lower latency for small files. No preprocessing needed.

Prerequisites

  • OS: Linux (tested on Ubuntu/Kubuntu), macOS, or Windows with WSL
  • Python: 3.8 or higher
  • Cloudflare Account: Free tier works
  • Domain: Any domain managed by Cloudflare
  • ChatGPT: Plus or Pro account (for MCP support)

Quick Start

1. Clone Repository

git clone https://github.com/YOUR_USERNAME/chatgpt-custom-mcp-for-local-files.git
cd chatgpt-custom-mcp-for-local-files

2. Install Dependencies

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

On Windows use: venv\Scripts\activate

3. Configure Environment

cp .env.example .env
nano .env

Edit these required variables:

BASE_DIR=/path/to/your/files
OAUTH_CLIENT_ID=chatgpt-mcp-default
OAUTH_CLIENT_SECRET=your-secure-random-string

Generate a secure secret:

openssl rand -hex 32

4. Set Up Cloudflare Tunnel

See detailed setup guide in docs/SETUP.md

Basic steps:

# Install cloudflared
curl -L https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb -o cloudflared.deb
sudo dpkg -i cloudflared.deb

# Create tunnel
cloudflared tunnel create mcp-files

# Configure DNS in Cloudflare dashboard
# Add CNAME: mcp.yourdomain.com -> TUNNEL_ID.cfargotunnel.com

# Start tunnel
cloudflared tunnel run mcp-files

5. Start MCP Server

python server_mcp.py

Server runs on http://localhost:4000

6. Connect ChatGPT

Go to ChatGPT → Settings → Apps and Connectors → Developer Mode → Enable

Then: Enabled Connectors → Create

Fill in:

Name: Local Files
URL: https://mcp.yourdomain.com
OAuth: Yes

ChatGPT will auto-register and authenticate.

Architecture

Simple request flow:

ChatGPT -> Cloudflare Tunnel (mcp.yourdomain.com) -> MCP Server (localhost:4000) -> Your Files

OAuth flow:

ChatGPT registers -> Gets client_id and client_secret -> Requests access token -> Uses token for API calls

Available Tools

ChatGPT can use these MCP tools:

list_files - List all files in directory

Example: "Show me all Python files"

read_file - Read complete file content

Example: "Read the README.md"

search - Search files by name

Example: "Find files containing 'config'"

fetch - Get file by path

Example: "Fetch src/main.py"

Security Considerations

What protects you:

  • OAuth 2.0 prevents unauthorized access
  • Files never leave your machine (served on-demand)
  • Cloudflare Tunnel encrypts all traffic
  • No credentials stored in ChatGPT

What you should do:

  • Only expose folders you want ChatGPT to access
  • Review BASE_DIR carefully before starting
  • Use strong OAuth credentials
  • Keep your tunnel credentials secure

See docs/SECURITY.md for detailed guidelines.

Troubleshooting

ChatGPT shows "424 Failed Dependency"

Check server logs for actual error. Verify OAuth token is valid. Ensure MCP response format is correct.

Tunnel connection fails

Verify DNS records point to tunnel. Check tunnel is running with cloudflared tunnel info. Review tunnel logs.

Server won't start

Check port 4000 is available with lsof -i :4000. Verify Python dependencies installed. Check .env file exists with correct values.

See docs/TROUBLESHOOTING.md for complete guide.

systemd Setup (Linux)

For automatic startup and management:

# Copy service files
cp setup/systemd/*.service ~/.config/systemd/user/

# Edit paths in service files
nano ~/.config/systemd/user/mcp-server.service

# Enable and start
systemctl --user daemon-reload
systemctl --user enable mcp-server mcp-tunnel
systemctl --user start mcp-server mcp-tunnel

# Check status
systemctl --user status mcp-server

Desktop Launchers (Linux)

# Copy desktop files
cp setup/desktop/*.desktop ~/.local/share/applications/

# Update desktop database
update-desktop-database ~/.local/share/applications/

Now "Start MCP Server" appears in your app menu.

Project Structure

.
├── server_mcp.py              Main MCP server
├── requirements.txt           Python dependencies
├── .env.example              Environment template
├── setup/
│   ├── tunnel-config.yml.example
│   └── systemd/
│       ├── mcp-server.service
│       └── mcp-tunnel.service
└── docs/
    ├── SETUP.md              Detailed setup guide
    ├── TROUBLESHOOTING.md    Common issues
    └── SECURITY.md           Security best practices

Contributing

Feel free to fork and modify for your own use. Pull requests are welcome but may not be reviewed or merged in a timely manner. No guarantees of acceptance.

License

MIT License - see LICENSE file for details.

Acknowledgments

Built following the MCP Protocol Specification at https://modelcontextprotocol.io/

Uses Cloudflare Tunnel for secure access: https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/

Implements RFC 7591 for dynamic client registration: https://tools.ietf.org/html/rfc7591

Support

Documentation: See docs/SETUP.md

Given as-is. No support provided.

Note: This is an independent project and is not officially affiliated with OpenAI or Anthropic.