Mastra

Mastra has no provider factory for this. The address is a field of the agent's model object: id and url.

npm i @mastra/core

import { Agent } from "@mastra/core/agent";

export const agent = new Agent({
  id: "researcher",
  name: "Researcher",
  instructions: "Answer precisely.",
  model: {
    id: "hanzo/zen5",
    url: "https://api.hanzo.ai/v1",
    apiKey: process.env.HANZO_API_KEY,
  },
});

Lands on POST /v1/chat/completions.

Two things the docs are explicit about. url must be the base URL of the OpenAI-compatible endpoint, never an individual chat endpoint. And id is a routing form: provider/model when the remote expects a bare model name, gateway/provider/model when the remote's own model namespace already carries the provider. Per-model headers are supported, with documented precedence over the client's.

MCP

MCPClient comes from @mastra/mcp. The URL is a url key holding a URL object, per named server.

import { MCPClient } from '@mastra/mcp';

export const mcp = new MCPClient({
  id: 'hanzo-mcp',
  servers: {
    hanzo: {
      url: new URL('https://api.hanzo.ai/v1/mcp'),
      requestInit: { headers: { Authorization: `Bearer ${process.env.HANZO_API_KEY}` } },
    },
  },
});

const tools = await mcp.listTools();

listToolsets() is the per-request variant, for when tools must not be shared across concurrent users.