I’ve been working on multi-agent AI systems (RAG pipelines, voice assistants, agent orchestration), and I’m curious how others in this community are connecting Directus to AI agent workflows, especially now that MCP (Model Context Protocol) is becoming a common way for agents to talk to structured data sources.
A few things I’d love to hear thoughts on:
- Has anyone exposed Directus collections to an AI agent via MCP or a custom tool layer?
- How are people handling permissions/scoping when an agent has read/write access to Directus content?
- Any experience using Directus as the “source of truth” for a RAG pipeline (e.g., syncing collections into a vector store)?
Directus feels like a natural fit as a structured backend for agentic apps, but I haven’t seen many concrete patterns discussed yet curious what others have tried.