Show HN: Share your AI Setup, Learn from others

Original Article ↗ Hacker News Discussion ↗

Developers are trading notes on a new site, mysetup.ai, that lets people share their AI tooling and workflows, from cloud agents and MCP servers to local LLM rigs and voice-driven setups. Many praise the idea of a centralized, searchable catalog of real-world configurations but push back on requirements like MCP integration, third‑party logins, and “explore with your agent” links they see as security risks. Underneath the tooling talk runs a deeper tension: whether to openly share increasingly powerful, customized AI workflows or treat them as proprietary advantages in an uncertain job market.

Overall reception

  • Many find the “share your AI setup” idea unexpectedly useful for discovering tools and practices they “didn’t know they didn’t know.”
  • Others feel setups are so individualized that copying them may be more confusing than helpful.

Contribution model, MCP, and security

  • Several object to requiring MCP and/or GitHub login to contribute; some explicitly refuse to connect arbitrary MCP servers or accounts.
  • The creator responds by adding more prominent manual entry, Markdown editing, image upload, and clarifying MCP is optional.
  • Some propose a local “collector” skill that users can inspect and run themselves to generate a standardized report.
  • Strong concern around pointing agents to unknown “explore this setup” endpoints and around agents self-adding MCP servers (“inherently insecure”).

Local and advanced AI setups

  • Multiple people share local setups: Qwen and Gemma variants running on consumer GPUs/CPUs, LM Studio, Subwave, EC2 + tmux, various MCP/skills stacks.
  • Several emphasize Tailscale-secured “homelab” networks and SSH workflows, including from phones.
  • Some run complex MCP ecosystems for code quality, de-duplication, ID handling, and delegating tasks to cheaper local models.

Voice and agentic workflows

  • One detailed account centers on low-latency voice-first work (custom ASR/TTS stack) and long-running agentic tasks with custom context-compaction.
  • Others struggle with ASR on technical terms and recommend specialized transcription tools with custom vocabularies.

Costs and subscription strategies

  • Concern about high token/API costs; interest in “approx monthly cost” fields and histograms of tool usage.
  • People mix subscriptions (Codex, Claude, Pi, etc.) with local models to offload cheaper or long-running work.

Features, UX, and data synthesis

  • Requests for: search/filter by tools and hardware, last-updated dates, popularity rankings, histograms of usage styles, RSS/API for agents to watch setups.
  • Some like the site design and responsiveness; others complain about initial lack of search and X-account framing.

Debate over sharing workflows

  • One camp views AI workflows as new proprietary “trade secrets” and advises against sharing, citing job insecurity and perceived exploitation by AI labs.
  • Another camp stresses that intuition and experience matter more than tools, sees value in open sharing, and notes that layoffs are often structural, not driven by individual productivity.