X Search One

Public product hub for the X Search MCP server: install, tools, security model, and an assistant that explains agent integrations.

Live: https://x-search-one.pages.dev

Public, no-account portfolio product in the same family as trip-one: grounded AI chat, light/dark UI, Cloudflare Pages, secrets server-side.

What it does

• Tool catalog — Documented MCP tools with inputs/outputs.
• Security model — Token handling, rate limits, least privilege.
• Install paths — pip, Docker, Railway — copy-paste ready.
• Agent chat — Ask how to wire search into an agent; answers cite the tool list.
• HTTP bridge — Call the same capabilities outside MCP clients.
• Portfolio-grade docs — Architecture diagram narrative recruiters skim in 60s.

Integrations

• Model Context Protocol — Tools for Claude, Cursor, and custom agents
• X API v2 — Bearer-authenticated search & user lookup
• Railway / Docker — Deploy the Python server anywhere
• Supabase optional — Cache or log query metadata

Engineering signals (for recruiters)

• AI tooling: MCP server design for external data
• API engineering: auth, rate limits, structured tools
• Python packaging + deploy (Docker/Railway)
• Developer experience: docs site + examples

Quick wins

• Interactive try-search playground with demo mode
• OpenAPI snippet generation
• Per-tool latency metrics panel
• Claude Desktop config generator

Stack

• Vite + React 18 + TypeScript (strict)
• React Router
• Cloudflare Pages + Functions (/api/chat, /api/health)
• OpenAI gpt-4o-mini (optional; UI works without it)

Develop

npm install
npm run dev

Copy .env.example to .dev.vars for Functions:

OPENAI_API_KEY=
AI_MODEL=gpt-4o-mini

Deploy

npm run build
npx wrangler pages deploy dist --project-name x-search-one --branch main

Set OPENAI_API_KEY on the Pages project for live chat.

git push updates GitHub only — deploy is a separate step.

Privacy

No accounts. Chat sends the on-page context + your message to /api/chat when AI is configured. No ads, no tracking pixels.

License

MIT