Welcome to Goa AI Discussions!

[raphael]

Welcome to Goa-AI Discussions! 👋

We're excited to have you here! This is the place to connect with the Goa-AI community—from newcomers exploring design-first AI backends to experienced developers building production MCP servers.

What to use Discussions for:

• 💬 Questions & Help: Ask about using Goa-AI, troubleshooting issues, or best practices for designing MCP servers
• 💡 Ideas & Feature Requests: Share ideas for new features, DSL improvements, or MCP protocol enhancements
• 🎨 Show & Tell: Share your Goa-AI projects, tools you've built, or interesting use cases
• 🐛 Bug Reports: While GitHub Issues are for confirmed bugs, Discussions are great for "is this a bug or am I doing something wrong?"
• 📚 Design Patterns: Discuss patterns for structuring tools, resources, prompts, and streaming responses
• 🤝 Integrations: Share how you've integrated Goa-AI with AI frameworks, agents, or other tools

Community Guidelines:

• Be welcoming and respectful—we're all learning
• Share code examples when asking for help
• Search existing discussions before creating a new one
• Keep discussions focused on Goa-AI, Goa framework, or MCP-related topics
• Remember: we're building this together

Get Started:

Drop a comment below and introduce yourself! Tell us:

• What brought you to Goa-AI?
• What are you building (or planning to build)?
• What AI/agent frameworks are you working with?

Looking forward to seeing what you build with Goa-AI! 🚀

---

For maintainers: Consider linking question-related issue templates to Discussions and converting old "question" or "discussion" labeled issues.

[raphael]

Thanks for the thoughtful introduction. Better Agent’s choice to preserve provider-native sessions rather than flatten them aligns well with Goa-AI’s separation of contracts from runtime adapters.

At the tool boundary, this works today: a Goa design can own stable input schemas, validation, typed results, and generated MCP adapters while Claude, Codex, and Gemini retain their own session and event models.

There is one important caveat around policy. Goa-AI generates tool tags and confirmation metadata and enforces allowlists and approval gates when its runtime owns execution. That policy metadata is not currently exposed as a portable envelope through MCP tools/list, and MCP itself does not standardize cross-client approval semantics. Native CLIs would therefore share the typed tools and results, but not automatically the same approval behavior.

The clean integration boundary seems to be: generated MCP services provide the canonical capability contract; Better Agent preserves each provider’s event stream; and one dispatch boundary enforces tool admission and approval before side effects. Exposing a typed projection of Goa-AI’s generated policy metadata for that boundary would be an interesting direction to explore. A concrete Better Agent integration would be very useful for shaping it.