Secure MCP Server Framework

What is this project?

Everybody wants to build with MCP, but no one is caring about the API Costs...

This is a secure, easy-to-use server that lets AI assistants like Claude or GPT safely access your tools, scripts, and APIs. It acts as a professional "security guard" between AI and your valuable resources, so you always stay in control.

---

Why did people need this?

Before:

• Many AI/automation tools could run your code, but anyone could connect, and there were no real safety checks built in.
• Most open MCP servers are like vending machines on the street: they let anyone press buttons, break things, or spend your money quickly. There's no login, no limit, no logs, and no safety net.

Problems this caused:

• People (or bots) could spam your tools, rack up big bills, or break important stuff.
• There was no way to see who did what, when, or why—so if anything went wrong, you couldn’t track it.
• Sensitive actions (deleting files, changing configs) could be done by anyone who found your server.

---

How does Secure MCP Server solve the problem?

This project adds multiple strong security layers:

Security Layer	What it does	How it works
Authentication	Only lets in people with correct username & password	Uses secure password hashing (bcrypt) and issues tokens after login
Authorization	Separates admin from regular users	Checks user role before letting them call sensitive tools
Rate Limiting	Stops spam and abuse	Automatically blocks tools after 60 requests per hour per user
Input Validation	Blocks dangerous commands from running	Code checks all inputs, blocks risky phrases (like 'delete files')
Audit Logging	Records who did what, when	Every tool use, error, and blocked action is written to a secure log
Sandboxing	Runs tools in isolated mode	Prevents tools from accessing files, system, or other risky resources
Token Checking	Requires every action to have a valid access token	Blocks any user or bot who tries to act without a real token

In short:

• ONLY the right person can use the right tool.
• Every attempt (good or bad) is logged.
• Rate limits stop spamming, cost overruns, and accidental infinite loops.
• Even if someone tries something risky, tools run in a secure sandbox that can't touch your system files.

---

How does it protect you compared to a basic MCP server?

	Basic MCP Server	Secure MCP Server
Login Required	❌ No (open to all)	✅ Yes (username & password)
Admin Permissions	❌ No	✅ Yes (only admin sees sensitive info)
Spam Protection	❌ No	✅ Yes (rate limits)
Blocks Dangerous	❌ No	✅ Yes (input check & sandbox)
Activity Logs	❌ No	✅ Yes (audit logging)
Knows Who Did What	❌ No	✅ Yes
Safe For Business	❌ No	✅ Yes

---

What technologies make this possible?

• Python 3.9+: Easy to read and fast to develop in.
• FastMCP & MCP SDK: Handles messages between AI and your server, so any AI app that uses MCP can connect.
• passlib + bcrypt: Keeps passwords safe so nobody can steal them.
• python-jose (JWT): Makes the 'ID cards' (tokens) that prove who each user is.
• SQLAlchemy + SQLite: Saves user, log, and tool data simply; easy to upgrade to PostgreSQL for big teams.
• structlog & prometheus: Simple, searchable logs and graphs to see how your system is running.
• psutil: Lets you monitor your own server’s CPU/memory usage.
• pydantic, python-dotenv: Clean, safe ways to use environment variables and settings.

---

Typical life of a request (How the layers protect you)

1. User logs in with username & password → gets a secure token.
2. Every time they do anything, the token is checked (like showing their ID card).
3. Before any tool runs, the server double-checks:
   • Are you allowed to do this? (admin vs. regular user)
   • Have you already used this tool too many times? (rate limiting)
   • Is your request safe? (input check)
   • Are you inside of your allowed memory? (context)
4. If all is good, the tool runs inside a sandbox (safe zone) so nothing dangerous can escape.
5. Every attempt (success or blocked) is written to the audit log for full traceability.
6. If anyone acts suspicious, they’re blocked and the admin can see everything in the logs.

---

Available Tools

Tool	Description
echo	Echo back text with its character length
calculator	Safely evaluate math expressions (2 ** 10, math.sqrt(144), etc.)
text_processor	uppercase, lowercase, title_case, reverse, word_count, char_count, strip
secure_hash	Generate MD5 / SHA-1 / SHA-256 / SHA-512 hashes
uuid_generator	Generate UUID v1 or v4
datetime_info	Current date/time in ISO, readable, timestamp, date-only, or time-only format
json_formatter	Pretty-print and validate JSON strings
base64_codec	Encode or decode Base64 strings

---

Use this MCP Server

There are two ways anyone can start using this server right now.

Option A: FastMCP Cloud (hosted, zero setup)

The server is deployed and live on FastMCP Cloud:

https://secure-mcp-server.fastmcp.app/mcp

Note: This server is hosted under the daredevil organisation on FastMCP Cloud. To connect, you need to be a member of the organisation.

How to join:

1. Go to fastmcp.cloud and sign in with your GitHub account.
2. Ask the maintainer (@dushyantzz) to invite you to the daredevil organisation — open an issue or reach out directly.
3. Once you accept the invite, you can connect immediately.

Connect from Claude Desktop — add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "secure-mcp-server": {
      "url": "https://secure-mcp-server.fastmcp.app/mcp"
    }
  }
}

Connect from Cursor — add this to your Cursor MCP settings:

{
  "mcpServers": {
    "secure-mcp-server": {
      "url": "https://secure-mcp-server.fastmcp.app/mcp"
    }
  }
}

---

Option B: Docker (self-hosted, no account needed)

Pull the image from DockerHub and run it yourself — no sign-ups, no org invites.

docker pull dushyantzz/secure-mcp-server:latest

Run the MCP server:

docker run -d -p 8000:8000 \
  -e SECRET_KEY=change-me-to-a-strong-secret \
  dushyantzz/secure-mcp-server:latest

Your MCP server is now running at http://localhost:8000/mcp.

Run the FastAPI REST server instead:

docker run -d -p 8000:8000 \
  -e SECRET_KEY=change-me-to-a-strong-secret \
  -e MODE=api \
  dushyantzz/secure-mcp-server:latest

REST API docs will be available at http://localhost:8000/docs.

Connect any MCP client to http://localhost:8000/mcp — same JSON config as above, just swap the URL.

With docker-compose (includes PostgreSQL, Redis, Prometheus, Grafana):

git clone https://github.com/dushyantzz/secure-mcp-server-framework.git
cd secure-mcp-server-framework
docker-compose up

---

Option C: Run from source

git clone https://github.com/dushyantzz/secure-mcp-server-framework.git
cd secure-mcp-server-framework

# Install dependencies
pip install -r requirements.txt

# Copy and edit environment config
cp .env.example .env

# Run the MCP server
fastmcp run server.py --transport streamable-http --port 8000

# Or run the FastAPI REST server
uvicorn mcp_server.main:app --host 0.0.0.0 --port 8000

---

Environment Variables

Variable	Required	Default	Description
SECRET_KEY	Yes	—	Secret key for JWT token signing
MODE	No	mcp	Docker mode: mcp (MCP server) or api (FastAPI REST server)
DATABASE_URL	No	sqlite+aiosqlite:///./secure_mcp.db	Database connection string
DEBUG	No	false	Enable debug mode and API docs
LOG_LEVEL	No	INFO	Logging level
RATE_LIMIT_RPM	No	60	Max requests per minute per user

See .env.example for the full list.

---

In summary

• Secure MCP Server gives you all the power of an MCP server
• PLUS enterprise-level safety and logs
• It adds a professional security guard, not just an open gateway
• Perfect for anyone who wants to run real workflows or tools with AI, safely