Server Tools & MCP Integration - Neosantara Docs

The Responses API (/v1/responses) supports two categories of tools: standard client-side function declarations and server-side tools orchestrated directly by the gateway via the Model Context Protocol (MCP) using the official OpenAI SDK.

Remote MCP Server Integration

By adding MCP server endpoints to the tools array, Neosantara operates as an MCP client and executes tools directly at the gateway level without requiring client-side execution loops.

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://api.neosantara.xyz/v1",
    api_key=os.environ["NEOSANTARA_API_KEY"]
)

# Attach remote MCP servers directly in tools array
response = client.responses.create(
    model="deepseek-v4.1-flash",
    input="Check priority bug issues in our repository.",
    tools=[
        {
            "type": "mcp",
            "server_label": "github",
            "server_url": "https://mcp.github-service.example.com/mcp",
            "headers": {
                "Authorization": "Bearer gh_mcp_token_secret"
            },
            "allowed_tools": ["list_issues", "get_issue"]
        }
    ]
)

print(response.output_text)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.neosantara.xyz/v1",
  apiKey: process.env.NEOSANTARA_API_KEY,
});

const response = await client.responses.create({
  model: "deepseek-v4.1-flash",
  input: "Check warehouse inventory stock for SKU-9821.",
  tools: [
    {
      type: "mcp",
      server_label: "inventory",
      server_url: "https://mcp.warehouse.example.com/mcp",
      headers: {
        Authorization: "Bearer warehouse_secret_token",
      },
    },
  ],
});

console.log(response.output_text);

Client-Side Function Declarations

When you want the model to generate structured arguments for functions executed on your own infrastructure, declare standard JSON function tools.

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://api.neosantara.xyz/v1",
    api_key=os.environ["NEOSANTARA_API_KEY"]
)

# Declare custom functions for model evaluation
response = client.responses.create(
    model="deepseek-v4.1-flash",
    input="What is the weather in Bandung today?",
    tools=[
        {
            "type": "function",
            "function": {
                "name": "get_weather",
                "description": "Fetch current weather by city",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "city": {"type": "string"}
                    },
                    "required": ["city"]
                }
            }
        }
    ]
)

print(response.output_text)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.neosantara.xyz/v1",
  apiKey: process.env.NEOSANTARA_API_KEY,
});

const response = await client.responses.create({
  model: "deepseek-v4.1-flash",
  input: "What is the weather in Bandung today?",
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        description: "Fetch current weather by city",
        parameters: {
          type: "object",
          properties: {
            city: { type: "string" },
          },
          required: ["city"],
        },
      },
    },
  ],
});

console.log(response.output_text);

Why Use Server MCP in Responses API?

In standard Chat Completions, when an AI model requests a tool call, your client application must catch tool_calls, execute code locally, and return output back to the gateway. With Responses API and type: "mcp":

Automated Gateway Execution: Neosantara connects to your remote MCP server directly.
Zero Client Orchestration: Your application simply awaits the finished result.
Async Synergy: Combines cleanly with background=True for long-running workflows.

Next Steps

Task	Guide
Full MCP Connector Guide	MCP Connector
Background Tasks	Background Tasks
Conversation State	Conversations & State

Related topics

OpenResponses API Function Calling (Tools)MCP Connector

⌘I