Integrations / For AI agents

Agents are consumers too.

My current R&D thesis: if AI agents write a growing share of UI code, design systems must be documented for agents as rigorously as for humans. This site practices that. A machine reading this page has three doors — and yes, the MCP endpoint is real.

fig. 01crawl

`/llms.txt`

A curated, plain-text summary of who I am, following the llms.txt convention. The polite front door for crawlers.

$ curl https://romaingabin.fr/llms.txt

fig. 02parse

`/resume.json`

My full profile in the JSON Resume schema. Same data that renders this website — one source of truth.

$ curl https://romaingabin.fr/resume.json

fig. 03connect — model context protocol

`/mcp`

A live MCP server (Streamable HTTP). Connect your agent and call get_profile, get_experience, get_projects or get_availability. Built the same way I built the MCP server for Carrefour’s design system — except this time, the component is me.

$ claude mcp add --transport http romain-gabin https://romaingabin.fr/mcp

Why bother?

Because assistants trained on the public web generate the public web’s average. At Carrefour I shipped an MCP server that fed agents our real design-system code — components, BEM conventions, WCAG rules — so “create a product card” produced production-ready code instead of generic markup. At Atos, I help product squads build the same muscle. A personal site that agents can query properly is the smallest honest demo of that idea.

hiring an agent to vet me? it can ask get_availability itself.