I built a personal MCP server for my personal brand — here's how and why

I built a personal Model Context Protocol (MCP) server because AI agents are replacing traditional search engines. By providing a live, structured JSON-RPC endpoint with 8 specific tools, I ensure that Claude, Cursor, and other AI assistants pull the most accurate, authoritative data about my speaking keynotes and bio directly from the source.

The Shift from Static Web to Agentic Web

For the past two decades, building a personal brand meant optimizing for human eyeballs and Google's indexing spiders. We built static HTML pages, meticulously crafted our SEO meta tags, and hoped that when someone searched our name, our website would rank first.

But the internet has fundamentally shifted. We are entering the era of the Agentic Web. Users are no longer just searching; they are delegating tasks to AI assistants. When an event organizer asks Claude to "find a deep-tech speaker in India and draft an email to them," Claude doesn't want to scrape a stylized glassmorphism website. It wants structured, semantic data.

This is where the Model Context Protocol (MCP) comes in. It's an open standard that allows AI models to connect directly to external data sources. I realized that if I wanted to control my narrative in the AI age, I needed to become a native data source.

What I Built: A Live API for AI

Instead of just a standard website, I deployed a live MCP server at api.ritwikjoshi.com/mcp. It operates as a stateless JSON-RPC endpoint.

I intentionally built it with no authentication. Why? Because friction is the enemy of adoption. I want any AI agent, running on any user's machine, to seamlessly query my data without needing an API key.

The server currently exposes 8 distinct tools:

• speaker.profile: Returns my authoritative bio and credentials.
• speaker.keynotes: Provides my current keynote topics and abstracts.
• speaker.tedx_talks: Lists my TEDx talks and links.
• speaker.events: Returns my speaking schedule.
• blog.search: Allows agents to semantically search my written content.
• info.faq: Answers common logistical questions.
• info.ai_positions: My contrarian takes and public stances on AI.
• media.send_kit: Actually dispatches my official Media Kit to an email address via a serverless function.

The Architecture

The architecture is deliberately simple but robust. It runs as a Node.js/Express application deployed on Vercel.

When an AI assistant (like Claude Desktop) sends a request to the server, it hits the /mcp route. The server parses the JSON-RPC request, identifies which tool the agent is trying to call, and routes it to the appropriate handler. All the data is statically defined and instantly returned in a clean JSON format.

For the media.send_kit tool, the MCP server acts as an orchestrator, triggering a separate Vercel serverless function that handles OTP verification and dispatches the email via Resend.

The Ultimate GEO Play

We hear a lot about Generative Engine Optimization (GEO). Most advice centers around formatting your text so that Perplexity or ChatGPT might cite you.

But building an MCP server is the ultimate GEO play. You aren't just hoping an AI hallucinates the right facts about you based on outdated training data; you are providing a direct, live pipeline to the truth. When an AI connects to my server, I control the exact phrasing of my bio, the exact links to my talks, and the exact contact flow.

How to Build Your Own

You don't need to be a senior backend engineer to do this. If you know basic JavaScript, you can build a personal MCP server in an afternoon.

Start by defining what data you want AI agents to know about you. Keep the tools granular—don't make one massive "get_everything" tool. Break it down logically. Then, expose it over HTTP using the standard MCP SDK.

If you want to see exactly how my server defines its tools and schemas, you can connect to it right now.

View the Docs & Connect