Lovable is expanding beyond AI-assisted app creation by letting builders expose parts of an application directly to AI agents. The company calls these reusable functions “capabilities” and publishes them as tools through a hosted Model Context Protocol server.
The result is one application with two entry points. People can continue using its graphical interface, while an agent in an MCP-compatible client such as ChatGPT or Claude can call approved functions without opening the app. A sales workflow, for example, could be invoked from a general work assistant rather than requiring someone to move between separate dashboards.
Chief technology officer Fabian Hedin describes the longer-term goal as a single interface connected to the tools and context a team uses every day. Lovable still intends to provide conventional app building and hosting; the agent layer extends those projects rather than replacing them outright.
The approach also creates governance questions. Builders must decide which functions an agent can access and what company context it should receive. Lovable says permission granularity and MCP security will be important as these capabilities move from demonstrations into workflows that can change business data.