How it works
The assistant connects to the MCP server, reads its list of tools, and decides when to call one based on what the user asks. The server runs the tool, for example a database query or an API call, and returns the result for the assistant to use in its answer.
Because tools run with the server's credentials, the server decides what an assistant is allowed to do. Good MCP servers expose narrow, well-described tools, require authentication, and make destructive actions explicit so a mistaken instruction cannot do much harm.
Example
A developer asks their assistant, "Give user 42 a 500-credit refund for yesterday's outage." The assistant calls the issue tool on the MCP server with user 42 and 500 credits, the server performs the API call, and the assistant reports the new balance of 1,340 credits back in the chat.
MCP server in Cred
The cred.diy MCP server lets an assistant manage credits, users, pricing, swaps, and referrals in your project, using your project's API key, so routine operations can be done from Claude, Cursor, or another MCP client.
MCP Server docsRelated terms
Questions
- Is an MCP server the same as an API?
- An MCP server usually wraps an API. The difference is that it describes its tools in a standard format AI assistants understand, so the assistant can choose and call them without custom integration code.
- Is it safe to let an assistant change balances?
- Only with care: scope the API key to the right project, review what tools are exposed, and use idempotency keys so a repeated tool call cannot move credits twice.
- Which assistants support MCP servers?
- The MCP project lists Claude, ChatGPT, Visual Studio Code, and Cursor among its clients. A client connects to a server and lists its tools for the model to use.
