What is MCP and why does this site support it?

MCP (Model Context Protocol) is an open standard that lets AI agents access external data sources in real time. This site exposes an MCP server so tools like Claude Code and Cursor can read and search everything here — live, not from training data.

How is this different from clicking Ask Claude or Ask ChatGPT?

Those buttons open a chatbot with a pre-typed query. The AI draws on training data — which may be months old and won't include recent posts. MCP gives your local AI agent direct read access to the actual content on this site, updated every time a new post or page goes live.

What can the AI access?

All blog posts and notes with full text, plus site pages including About, Work history, Uses This, Side Quests, Fun, and Changelog. Six tools: get_site_info, list_posts, get_post, list_pages, get_page, and search.

Does ChatGPT support MCP?

Not yet via the web interface. OpenAI's Codex CLI does support MCP. Check the Codex tab on this page for setup instructions.

Connect via MCP

Q: How do I connect?

Add the MCP server URL to your AI client. For Claude Code: claude mcp add --transport http varunchoraria https://varunchoraria-mcpvercelapp.vercel.app. For Claude Desktop or Cursor, add the URL to your mcpServers config JSON.

Let your AI read this site — posts, pages, everything — live, not from training data.

https://varunchoraria-mcpvercelapp.vercel.app

Paste this URL into any MCP-compatible AI client.

Connect by client

Run this once in your terminal:

claude mcp add --transport http varunchoraria https://varunchoraria-mcpvercelapp.vercel.app

Done. Open a new Claude Code session and ask anything.

Open Settings → Developer → Edit Config and add inside mcpServers:

"varunchoraria": {
  "type": "http",
  "url": "https://varunchoraria-mcpvercelapp.vercel.app"
}

Restart Claude Desktop after saving.

Go to Settings → MCP and add a new server, or edit .cursor/mcp.json:

"varunchoraria": {
  "type": "http",
  "url": "https://varunchoraria-mcpvercelapp.vercel.app"
}

Reload Cursor after saving.

Add to ~/.codex/config.yaml:

mcpServers:
  varunchoraria:
    type: http
    url: https://varunchoraria-mcpvercelapp.vercel.app

No restart needed — Codex CLI picks it up on next run.

ChatGPT's web interface doesn't support MCP yet. Options:

Codex CLI — OpenAI's terminal tool does support MCP. See the Codex tab above.
Claude Code or Cursor — both work with this server today.
Manual — copy a post from the archive and paste it into ChatGPT as context.

MCP in ChatGPT is on OpenAI's roadmap. Check back when it ships.

Any MCP-compatible client that supports HTTP transport works. Point it at:

https://varunchoraria-mcpvercelapp.vercel.app

Transport: HTTP (MCP spec 2024-11-05). No auth required.

What to ask once connected

"What's Varun's take on AI-first GTM strategy?"
"Summarise his notes on management and team leadership."
"What tools and software does he use day to day?"
"What's his most recent post about?"
"What side projects has he shipped with AI?"

See it in action

Real Claude Code sessions using the MCP server — no copy-pasting, no manual context.

Claude Code using the varunchoraria MCP server to pull the top 3 insights from career blog posts — *"Use varunchoraria MCP and tell me the top 3 insights across his career blog posts."* Claude fetched the posts directly and surfaced three sharp takeaways — including "Luck = Do × Tell (50:50)" from I've been trying to get lucky for a decade.

Claude Code using the varunchoraria MCP server to summarise career trajectory from work history page — *"Where does he work, and summarise his career trajectory for me."* Claude pulled the Work page and mapped the full arc: student founder → founding PMM → PLG operator → category-creating marketing manager.

How this is different from Ask Claude / Ask ChatGPT

The "Ask Claude" and "Ask ChatGPT" buttons on this site open a web chatbot with a pre-typed query. That's useful for a quick introduction — but those AIs draw on training data, which has a cutoff date and doesn't include recent posts or pages added after their last training run.

MCP is for a different use case. When you wire up an AI agent in Cursor or Claude Code, it gets live read access to everything on this site — content published today included. Instead of the AI guessing at what I've written, it reads the actual text directly. The result is more accurate, more recent, and fully cited.

Use the chatbot buttons for discovery. Use MCP when you're doing real work — building something, writing a brief, doing research — and want my writing as active context in your workflow, not a vague memory from training.

Why this exists

Personal websites are dark matter to AI. Most models have a knowledge cutoff, don't index small personal sites deeply, and go stale fast. Ask a chatbot what I've written about B2B go-to-market strategy and it'll give you a generic answer — not the specific frameworks and cases I've actually published.

MCP is the protocol that fixes this. By running a small server that speaks MCP, any compatible AI agent can access what's here in real time. No copy-pasting, no RAG pipeline, no stale embeddings. The content updates automatically whenever I push new writing — the server always reads from the live site.

This is also an experiment. I'm interested in what it looks like when a personal site becomes a first-class data source for AI agents, not just a URL to paste into a chat window.

What the AI can access

get_site_info Bio, role, topics, contact links

list_posts All notes — titles, dates, tags

get_post Full text of any post by slug

list_pages About, Work, Uses This, and more

get_page Full content of any site page

search Keyword search across everything