This is a live demonstration of an AI agent (GLM-4.5) interacting with Rynko Flow via the Model Context Protocol (MCP).
What you are watching:
The Failure: The agent submits an invalid JSON payload to a Rynko Flow gate.
The Rejection: Instead of silently failing, the Rynko MCP server evaluates the payload against strict schemas and cross-field business rules, returning a structured JSON-RPC error detailing exactly what failed (e.g., amount must be > 0, currency enum mismatch).
The Self-Correction: Without any custom Python retry loops or LangGraph interrupt nodes, the agent reads the structured error, reasons about the constraints, and autonomously generates a corrected payload.
The Success: The second attempt passes the gate and returns a tamper-proof validation_id.
Why this matters: This proves you can move business validation out of your orchestrator code and into a deterministic, external control plane.
Read the full architectural breakdown and test report here: https://blog.rynko.dev/launching-rynko-flow-a-validation-gateway-for-ai-agent-outputs