@mcpc-tech/aiyo-sampling

MCP Sampling integration package for @mcpc-tech/aiyo.

Use this package when you want the adapter to route LLM calls through an MCP Sampling provider — i.e. the connected MCP client (such as VS Code Copilot) handles the actual model call.

Install

pnpm add @mcpc-tech/aiyo-sampling

Quick start

As a library

import { Hono } from "hono";
import { createAiyo } from "@mcpc-tech/aiyo-sampling";

const adapter = createAiyo({
  defaultModel: "copilot/gpt-4o-mini",
  defaultSamplingConfig: { server },
});

const app = new Hono();
app.get("/v1/models", adapter.honoHandler());
app.post("/v1/chat/completions", adapter.honoHandler());
app.post("/v1/responses", adapter.honoHandler());
app.post("/v1/messages", adapter.honoHandler());

As a stdio MCP server

npx aiyo-sampling

This starts:

1. An MCP server on stdio — exposes an ask_ai tool that the connected MCP client can invoke.
2. An HTTP server on port 3456 (configurable via PORT env var) — exposes OpenAI-compatible endpoints.

MCP client config (e.g. VS Code settings.json)

{
  "mcp.servers": {
    "sampling-aiyo": {
      "command": "npx",
      "args": ["@mcpc-tech/aiyo-sampling"]
    }
  }
}

What this package adds

Compared with the core package, this package wires in:

• MCP Sampling-backed runtimeFactory
• MCP Sampling-backed listModels

Helper exports

This package also exports these helpers when you want lower-level control:

• createSamplingRuntimeFactory
• createSamplingListModelsResolver

It re-exports the core adapter types and helpers from @mcpc-tech/aiyo.