MCP (Model Context Protocol)

What MCP Is

MCP is an open protocol that lets AI applications connect to external tools and data through standardized MCP servers — file systems, databases, search, internal APIs, and more. A client that speaks MCP can discover the tools an MCP server exposes and let the model call them during a conversation.

How MCP Works with MyTokenGate

MCP has two moving parts:

• The MCP servers, which expose tools and data. They run inside your own client environment.
• The model, which decides when to call those tools.

MyTokenGate is the model backend, not an MCP host. You configure MCP servers in your client (Claude Code, Cursor, Cline, and others) following that client’s own MCP documentation, and point the client’s model endpoint at the MyTokenGate gateway. During a session:

1. Your client sends the conversation plus the MCP servers’ tool definitions to MyTokenGate as a normal tool-calling request.
2. A capable model returns tool calls (see Function Calling).
3. Your client executes those calls against the local MCP servers and returns the results.
4. Tool execution stays on your side; MyTokenGate only forwards inference.

Because MCP tool calls travel as standard function-calling requests, any model that supports function calling works with MCP through the gateway.

Setup

1. Configure MyTokenGate as the model backend in your client — see Claude Code, Cursor, or Cline.
2. Add MCP servers in the client following its own MCP documentation.
3. Use a model that supports function calling.

Notes

• MyTokenGate does not host or proxy MCP servers; they run in your client.
• MCP tool execution, and any data those tools touch, stay within your environment.
• A model’s MCP support is the same as its function-calling support.