kranners/jailbreak-mcp ? reverse-engineered prompt

Reverse engineered prompt

Build me a TypeScript MCP middleware server that sits between my AI client and my other MCP servers, and lets me control which tools are available so I don’t blow up the context window.

I want it to read a config file from either a path argument or an environment variable, validate it, connect to multiple MCP servers, and then expose a smaller curated set of tools through one single server. It should support different modes, so I can switch between things like a normal mode, admin mode, or focused task specific mode, and each mode can allow all tools from a server or only selected ones.

If possible, also make it work around clients that only understand tools by turning prompts and resources into tools too. Keep the setup simple, use the config format shown in the README, and make it easy to run with Cursor or similar clients. If you need current MCP details, look them up online first.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab