Adding a Model Context Protocol server to your developer tool site

The Model Context Protocol (MCP) lets you expose tools to AI agents in a standard way. Once you publish an MCP endpoint, any compatible client — Claude Code, Cursor, Cline, and a growing list of others — can call your tools without per-client integration work.

We exposed 10 of convertifydata's converters as an MCP server. The whole thing is a single Netlify Function. Here is how the plumbing works and why it is worth doing for any utility site.

The shape of an MCP server

At the protocol level, an MCP server is an HTTP endpoint that speaks JSON-RPC. Clients send requests like tools/list to enumerate available capabilities and tools/callto invoke one. The server registers tools by name and provides a Zod-style schema for each tool's arguments.

The Anthropic SDK does most of the work:

const server = new McpServer({ name: "convertifydata", version: "1.0.0" });

server.tool(
  "json_to_yaml",
  "Convert JSON to YAML",
  { json: z.string(), indent: z.number().optional() },
  ({ json, indent }) => ({
    content: [{ type: "text", text: jsonToYaml(json, indent ?? 2) }],
  })
);

That is the full registration for one tool. The function body — the actual conversion — is the same code that runs in the browser. We export each converter from a shared file and use it from both the browser bundle and the function.

Why Netlify Functions specifically

The site is a fully static export. Adding one serverless function for the MCP endpoint kept the rest of the architecture untouched and meant no separate hosting bill. Cold starts on Netlify are ~200ms which is well below the threshold AI clients consider "snappy".

Why this matters for traffic

Every developer who installs the MCP server becomes a recurring, invisible user. They will never visit the homepage again, but they will cite the URL when their agent uses one of the tools. We see install traffic growing 3-5x faster than browser traffic — the asymmetry is real and only widens as AI coding workflows mature.

How to install it

claude mcp add convertifydata -- npx -y mcp-remote https://convertifydata.com/mcp

That one line, and your agent has 10 conversion tools available. The full list is documented on the MCP server page.

Worth doing if you run a utility site

Exposing your existing tools through MCP took us a few hours and added zero ongoing maintenance — the function shares the same source files as the browser tools. If your site does anything an AI agent might want to call (formatting, transcoding, validating, calculating), the upside of being the default MCP endpoint for that capability is large and growing.