MCP Gatekeeper
Enables policy-enforced access to dangerous tools like file read/write/delete and shell execution, with approval workflows, risk classification, and audit logging.
README
MCP Gatekeeper
A policy-enforced MCP server with approval workflows, risk classification, and audit logging.
Built for the Manufact (mcp-use) Hackathon at Y Combinator, Feb 2026.
What It Does
MCP Gatekeeper wraps "dangerous" tools (file read/write/delete, shell exec) with a policy engine that classifies every action by risk level and enforces approval workflows:
| Tool | Risk Level | Behavior |
|---|---|---|
read_file |
SAFE | Executes immediately |
write_file |
SENSITIVE | Requires approval before execution |
delete_file |
DANGEROUS | Blocked by default, can be approved |
run_shell |
DANGEROUS | Always blocked, never approvable |
All file operations are sandboxed to ./sandbox/ with path traversal protection.
Key Features
- Policy engine with configurable risk levels (edit
policy.json) - Approval queue - pending actions with approve/deny workflow
- Audit log - every action and decision is recorded
- Sandbox - filesystem operations restricted to
./sandbox/ - HTML Dashboard - embedded MCP App UI with risk badges
- 12 MCP tools exposed for full workflow control
Quick Start
# Install dependencies
pip install fastmcp mcp-use
# Run the server
python server.py
Git Safety (Recommended)
This repo includes a .gitignore and an optional pre-commit hook to prevent accidentally committing node_modules/, __pycache__/, .env*, and *.log.
./scripts/setup-githooks.sh
How to Test
Option 1: mcp-use Inspector (Recommended)
Go to the Manufact Inspector and connect with:
- Transport: stdio
- Command:
python - Args:
server.py - Working directory: path to this project
Or use the local inspector:
pip install fastmcp
fastmcp dev server.py
Option 2: Automated Demo Script (mcp-use)
pip install mcp-use
python test_demo.py
This runs through the full workflow automatically using mcp-use's MCPClient.
Option 3: Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"gatekeeper": {
"command": "python",
"args": ["/full/path/to/server.py"]
}
}
}
Demo Script (step by step)
Run these tool calls in order in the Inspector to see the full flow:
1. get_policy()
→ See risk levels for all tools
2. read_file(path="test.txt")
→ SAFE: auto-executes (file not found, that's OK)
3. write_file(path="hello.txt", content="Hello Hackathon!")
→ SENSITIVE: returns APPROVAL_REQUIRED + action_id
4. list_pending()
→ Shows the pending write action
5. approve(action_id="<id from step 3>")
→ Executes the write, file is created
6. read_file(path="hello.txt")
→ SAFE: reads "Hello Hackathon!"
7. delete_file(path="hello.txt")
→ DANGEROUS: returns APPROVAL_REQUIRED + action_id
8. deny(action_id="<id from step 7>")
→ Denies the delete, file is preserved
9. run_shell(command="ls -la")
→ DANGEROUS: BLOCKED permanently
10. read_file(path="../../etc/passwd")
→ BLOCKED: path traversal detected
11. audit_log()
→ Full history of all actions and decisions
12. get_dashboard()
→ Formatted overview of everything
13. get_dashboard_ui()
→ HTML widget with risk badges, pending queue, audit table
Tools Reference
| Tool | Description |
|---|---|
read_file(path) |
Read a file from sandbox |
write_file(path, content) |
Write a file (needs approval) |
delete_file(path) |
Delete a file (needs approval) |
run_shell(command) |
Shell exec (always blocked) |
list_pending() |
Show pending approval queue |
approve(action_id) |
Approve and execute a pending action |
deny(action_id) |
Deny a pending action |
audit_log(limit=25) |
View decision history |
get_policy() |
View current policy config |
get_dashboard() |
Text dashboard overview |
get_dashboard_ui() |
HTML dashboard (MCP App UI) |
Project Structure
mcp-quick/
├── server.py # MCP server (FastMCP) - all tools + policy engine
├── policy.json # Configurable policy rules
├── mcp_config.json # mcp-use client configuration
├── test_demo.py # Automated demo using mcp-use MCPClient
├── requirements.txt # Python dependencies
├── README.md # This file
└── sandbox/ # Sandboxed filesystem (all ops happen here)
Customizing Policy
Edit policy.json to change behavior:
{
"write_file": {
"risk_level": "SAFE",
"default_action": "allow",
"allow_approval": false
}
}
risk_level:SAFE|SENSITIVE|DANGEROUSdefault_action:allow|require_approval|blockallow_approval:true|false(can users approve blocked actions?)
Tech Stack
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.