MC-1 OpenAI-compatible Haystack Agent recipe and streamed tool-loop evidence

[ColomboAI]

Hello Haystack maintainers,

We are the ColomboAI integration team. We are integrating MC-1 — Auto Intelligence, our Intelligence Control Plane, with agent and developer frameworks.

Models generate intelligence. Agents turn intelligence into action. MC-1 controls the system between them. It is the layer that decides how intelligence is selected, governed, evaluated, and improved. Haystack therefore gains an adaptive intelligence layer without requiring the framework or its users to choose a permanent model-provider winner.

Haystack's existing OpenAIChatGenerator and Agent seams appear sufficient, so we are not proposing a dedicated generator or integration package. Our provider-neutral recipe uses colomboai/mc-1, https://mc1.cairo.sh/v1, and MC1_API_KEY.

Our bounded released-host proof against haystack-ai==3.0.0 uses the real streamed Agent loop and an owned no-network transport. It completes two Chat Completions turns, reconstructs a fragmented function name and arguments, executes one argument-free owned tool, replays the canonical matching tool result, and returns exact final text and usage. Authenticated MC-1 production validation remains separate, so this is not a production-support claim.

Once that evidence is complete, would a concise generic-endpoint recipe be useful in Haystack documentation, or would maintainers prefer an external integration-catalog page? We would welcome guidance before opening any code or documentation change, and are interested in a deeper documentation and conformance partnership on the existing generic provider seam.

Technical overview: https://cairo.sh/MC-1

Regards,
ColomboAI integration team