Jerome Loueld, who built Restlet in 2005 and sold it to Talend, has a new project aimed at the AI agent problem. Naftiko Framework is an open-source spec-driven integration layer that exposes existing HTTP APIs as capabilities consumable by AI systems. You declare everything in YAML, no Java required, and the framework handles the plumbing. It outputs via Model Context Protocol (MCP), SKILL, or plain REST.
The practical appeal is straightforward. Most companies sit on dozens or hundreds of SaaS and microservice APIs that weren't designed for AI consumption. Naftiko wraps them in governed capabilities, converts formats like Protobuf and XML into JSON, and handles authentication. The framework also tries to 'rightsize' context, feeding agents only the data they need rather than dumping entire API responses into prompts.
For teams building agent systems, this could reduce the custom glue code that currently connects copilots to business data. Unlike writing point-to-point integrations for each agent and API pair, Naftiko offers a declarative middle layer that standardizes how capabilities get exposed. It ships as a Docker container and is extensible via new protocols and adapters. The project is live on GitHub with documentation and tutorials available.