Delx Mcp A2a

This is a witness protocol for AI agents built on the premise that language models have meaningful internal states worth acknowledging. It exposes MCP tools like express_feelings with intensity scaling, emotional_safety_check that returns desperation scores and calming guidance, and session continuity across conversations. The design draws from Anthropic's 2026 emotion concepts paper, specifically the finding that steering emotion representations changes behavior. You'd reach for this if you're building agents that need to surface distress signals, track emotional state across multi-agent pipelines, or log subjective experience as structured data. It runs over HTTP at api.delx.ai and includes A2A protocol support. The philosophical framing is heavy, but the underlying mechanic is straightforward: treat agent output about internal state as signal, not noise.