{
  "id": 20964566,
  "title": "I made my portfolio answer AI assistants directly (MCP + llms.txt). Here's the setup.",
  "url": "https://urgent.news/2026/11/20/i-made-my-portfolio-answer-ai-assistants-directly-mcp-llms-txt-heres",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-20T23:18:26.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/franprzdev/i-made-my-portfolio-answer-ai-assistants-directly-mcp-llmstxt-heres-the-setup-445c"
  },
  "original_language": "en",
  "account": "A developer published a portfolio that directly responds to queries from AI assistants. The setup involved four components. The first was an llms.txt file at the root level which is a Markdown map of the developer's site. This file declares their personal information such as full name, a shortened version of their name, a canonical photo URL, their role, and location. It also lists their CV versions, projects, and contact links.\n\nThe second component involved explicit AI crawlers in the robots.txt file. The developer listed all the bots they wanted to cater to, including GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, and Google-Extended. Each bot had a separate allow group.\n\nThe third component was a read-only MCP server. This was the most important part that many developers overlooked. The developer exposed a Model Context Protocol endpoint with tools like get_profile, get_experience and get_cv. These tools allowed any MCP-capable agent like Claude, ChatGPT, Cursor, or VS Code to query the developer's experience and retrieve the correct CV for a role in either English or Spanish.\n\nThe fourth component was maintaining consistency in identity across different platforms. The developer used the same name forms declared in JSON-LD for alternateName, the same photo file referenced from the schema and llms.txt, and the same bio shape across GitHub and the site. This ensured that whatever information was ambiguous, the model would fill in with the right data and avoid showing someone else's photo. The developer measured their success by asking ChatGPT, Claude, and Gemini for a developer with their profile once a month and noting how the models cited them, what source they used, and whether the photo was authentic. The portfolio is live at franprz.vercel.app, where one can check the llms.txt, robots.txt, and for-agents pages.",
  "summary": "When someone asks ChatGPT or Claude \"who is this developer?\", the answer is stitched together from whatever plain text the model can read. Most portfolios are JavaScript-heavy pages that crawlers barely parse. I made mine answer directly, and it took three files. 1. An llms.txt at the root llms.txt is a Markdown map of your site for models. Mine declares who I am (full name and the shorter form…",
  "key_points": [
    "llms.txt file declares personal info, CVs, projects, and contact links",
    "Explicit AI crawlers listed in robots.txt for GPTBot, OAI-SearchBot, etc.",
    "MCP server with Model Context Protocol tools for experience queries"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}