Assistants & MCP

Grok Bot & xAI MCP Integration

Equip Grok bots (Grok 3, Grok 3 Mini, Grok 2 Vision) with Model Context Protocol (MCP) tool calling — executing live GPU queries, multi-model completions, and image generation via OneInfer's MCP server.

Connect Grok bots to OneInfer using the streamable MCP endpoint https://mcp.oneinfer.ai/mcp. The Grok agent dynamically negotiates tools for model queries, pricing, image generation, and GPU provisioning.

Grok MCP Architecture & Setup

Rather than hardcoding REST calls, configure the Grok agent runner with OneInfer's MCP endpoint.

Configure your bot agent or IDE to communicate with OneInfer via MCP:

{
  "mcpServers": {
    "oneinfer": {
      "type": "streamable-http",
      "url": "https://mcp.oneinfer.ai/mcp"
    }
  }
}

The Grok reasoning model accesses tools dynamically over HTTP/SSE without hardcoded API endpoints.

Configuring Grok Bot with OneInfer MCP

Add OneInfer to your bot or agent's MCP configuration manifest:

json

{
  "mcpServers": {
    "oneinfer": {
      "type": "streamable-http",
      "url": "https://mcp.oneinfer.ai/mcp"
    }
  }
}

Executing MCP Tools from Grok 3

When running Grok 3 in an MCP-enabled agent harness, the model selects and executes OneInfer tools natively:

python

from mcp.client.sse import sse_client
from mcp import ClientSession

async with sse_client("https://mcp.oneinfer.ai/mcp") as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        
        # Grok bot asks for available models via MCP
        result = await session.call_tool(
            "oneinfer_list_models",
            arguments={}
        )
        print("Available models:", result.content)