Scope an MCP Server for This Codebase
Published · Updated
Works out which tools an MCP server should expose before you write any of it.
Tags: AI, Engineering, Planning
Prompt template
## Before I build the MCP server I want an agent to be able to work against this system. Help me decide what the server should expose rather than wrapping every endpoint I happen to have. - [textarea: What the system does and who uses it] - [textarea: The API surface or main entry points available today] - [dropdown: Read only, Read and write, Write with confirmation on destructive actions] - [text: What an agent using this should be able to accomplish end to end] - [toggle: Yes, No] There is data here that must never leave the environment Propose a tool list: name, one-line description, arguments, and what it returns. Keep the count small and make each tool map to something a caller actually wants to do, not to a database table. Point out where two of my endpoints should collapse into one tool, and where one endpoint needs splitting because the agent will otherwise have to guess at a mode flag. Then tell me which tools need a confirmation step and what the failure text should say, since that is what the model sees when it goes wrong.