What is Cordon?
Cordon is an open-source security gateway designed to control and monitor calls to Model Context Protocol (MCP) tools. It sits between AI applications and the tools they access, allowing you to inspect, filter, and approve tool invocations before they execute. This is particularly useful when running AI agents that need access to external services or sensitive operations. Cordon helps prevent unauthorised or unintended tool usage by giving you visibility and control over what your AI systems can do. The tool is self-hosted, meaning you run it on your own infrastructure rather than relying on a third-party service.
Key Features
Tool call inspection
View details of tool invocations before execution to understand what your AI system is attempting to do
Access control
Define policies to restrict which tools can be called and under what conditions
Request filtering
Block or modify tool requests based on custom rules and parameters
Audit logging
Track all tool calls and decisions for compliance and debugging purposes
Open-source codebase
Full transparency and ability to customise the gateway to your specific needs
MCP compatibility
Works directly with Model Context Protocol implementations
Pros & Cons
Advantages
- Complete control over tool access without relying on external security providers
- Free to use and modify since it is open-source software
- Transparent inspection of AI agent behaviour helps identify unwanted or dangerous actions
- Self-hosted deployment means your tool call data stays within your infrastructure
Limitations
- Requires technical knowledge to set up and maintain as a self-hosted solution
- You are responsible for managing security updates and infrastructure reliability
- Limited built-in integrations; you may need to write custom policies for your specific use case
Use Cases
Running AI agents in production environments where you need to audit every external action they take
Controlling which APIs or services an AI application is permitted to call
Implementing approval workflows for sensitive tool operations before they execute
Monitoring and logging tool usage for security compliance and troubleshooting