Can TCA be used to model "skills" (MCP / agent capabilities) for AI-driven workflows?

[asyncawaited]

Hi everyone,

I'm exploring a direction that combines TCA with AI-driven development workflows (e.g. vibe coding, MCP, Codex-like agents), and I’m curious if anyone has experimented with something similar.

In these systems, we often define "skills" (or MCP capabilities), which are:

• Composable units of behavior
• Can be orchestrated dynamically by an agent
• Often involve async effects (networking, IO, device interaction)
• Require clear state transitions, cancellation, and observability

From my perspective, TCA seems like it could be a very strong foundation for modeling this:

• Reducers → skill state transitions
• Effects → async skill execution
• Store → runtime orchestration / coordination
• Dependency system → injectable capabilities

So my questions are:

1. Has anyone tried modeling "skills" or MCP-style capabilities using TCA?
2. Do you think TCA is suitable as a runtime for orchestrating AI-driven workflows?
3. Are there any patterns or limitations I should be aware of?

I’m especially interested in:

• dynamic composition of skills
• long-running effects and cancellation
• external orchestration (e.g. an AI deciding which reducer/feature to trigger)

Would love to hear thoughts from the community 🙏

[Alex293]

https://www.pointfree.co/the-way