AI Tool & MCP Server Generator

How it works

1. Paste a curl command: from the API's docs, your terminal, or your browser's dev tools (or convert one with the cURL converter).
2. The generator turns path IDs, query parameters and JSON or form body fields into tool parameters, with types inferred from your example values.
3. Rename the tool and improve the descriptions: that's what the model reads when deciding what to call.
4. Copy the format you need: Claude, OpenAI, an MCP server, or just the handler.

Keep API keys away from the model

A model should never see your credentials: anything in its context can end up in its output. The generator spots Authorization and API-key headers, -u credentials and key-like query parameters (access_key, apikey, appid, token…), removes them from the schema and reads them from environment variables in the handler instead.

Use it with Claude

Pass the definition in tools. When Claude stops with stop_reason: "tool_use", run the handler for each tool_use block and send the results back as tool_result blocks, all in one message:

import Anthropic from "@anthropic-ai/sdk";
import { getForecast } from "./handler.js"; // the generated handler

const client = new Anthropic(); // reads ANTHROPIC_API_KEY

// Paste the "Claude (Anthropic)" definition from the generator here.
const tools: Anthropic.Beta.BetaTool[] = [/* … */];

const messages: Anthropic.Beta.BetaMessageParam[] = [
  { role: "user", content: "Do I need an umbrella in Lagos today?" },
];

while (true) {
  const response = await client.beta.messages.create({
    model: "claude-opus-5-5",
    max_tokens: 16000,
    tools,
    messages,
    // If a request is declined, retry it on a fallback model automatically.
    betas: ["server-side-fallback-2026-07-01"],
    fallbacks: "default",
  });
  messages.push({ role: "assistant", content: response.content });

  if (response.stop_reason !== "tool_use") {
    for (const block of response.content) if (block.type === "text") console.log(block.text);
    break;
  }

  // Run every tool call, then send all results back in one message.
  const results: Anthropic.Beta.BetaToolResultBlockParam[] = [];
  for (const block of response.content) {
    if (block.type !== "tool_use") continue;
    try {
      results.push({ type: "tool_result", tool_use_id: block.id, content: await getForecast(block.input) });
    } catch (error) {
      results.push({ type: "tool_result", tool_use_id: block.id, content: String(error), is_error: true });
    }
  }
  messages.push({ role: "user", content: results });
}

Use it with OpenAI

Pass the definition in the tools array. With the Responses API, each tool call arrives as a function_call item: run the handler with JSON.parse(item.arguments) and send back a function_call_output item with the same call_id. With Chat Completions, use the nested { type: "function", function: {...} } format and reply with a tool message.

Use it as an MCP server

Save the MCP server code as server.ts, install @modelcontextprotocol/sdk and zod, and add it to your MCP client. For Claude Desktop, add it to claude_desktop_config.json:

{
  "mcpServers": {
    "get-forecast": {
      "command": "npx",
      "args": ["tsx", "/absolute/path/to/server.ts"],
      "env": { "OPEN_METEO_API_KEY": "only-if-the-api-needs-one" }
    }
  }
}

In Claude Code: claude mcp add get-forecast -- npx tsx /absolute/path/to/server.ts. Browse existing MCP servers in the directory too.

Free APIs that make great AI tools

• Open-Meteo: weather forecasts, no key.
• REST Countries: country facts, currencies and languages.
• Nager.Date: public holidays for 100+ countries.
• Frankfurter: currency exchange rates.
• ipstack: IP geolocation, so an agent can localise answers (free plan with a key).

Find more in the APIs with no key list, and generate schemas for structured outputs with the JSON Schema generator.