Function Calling (OpenAI)

The complexity of parallel function execution

Function calling—introduced by OpenAI—allows developers to pass an array of available tools in the tools array of a chat completion request. The model then intelligently decides which function to invoke, providing the necessary JSON arguments.

However, handling the lifecycle of these calls is difficult. Models now support parallel tool calls, meaning the model might ask to execute three functions simultaneously. Your infrastructure must securely hold credentials, execute the network requests, and manage retries. wmcp.sh acts as an execution gateway that handles this automatically, utilizing an encrypted credentials vault for static keys to keep your environment secure. (Note: wmcp.sh is not affiliated with OpenAI, Anthropic, or any other mentioned orgs.)

Executing a function call with OpenAI

import OpenAI from "openai";

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

// Fetch dynamic tool schemas from wmcp.sh rather than hardcoding
const toolSchemas = await fetch("https://api.wmcp.sh/v1/tools?provider=openai").then(r => r.json());

const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Fetch the latest logs." }],
  tools: toolSchemas,
  tool_choice: "auto"
});

// Inspect the model's decision
const toolCalls = response.choices[0].message.tool_calls;
if (toolCalls) {
  for (const call of toolCalls) {
    console.log("OpenAI requested function:", call.function.name);
    // In production, wmcp.sh executes these in parallel
  }
}

Managing Function Calls at Scale

Common questions.