gloria.dev works as one loop. Your agent’s skills read the project’s own source, the results move into gloria.dev, and the service keeps the picture current as the code changes. That is how the product keeps a project aligned with what the code actually does.
If the terms are new, start with Concepts for the shared vocabulary behind projects, skills, and MCP.
Three parts, one loop
Coding-agent skills are the reading layer. They run inside your agent and inspect the project’s source, config, and docs structure directly. Their job is to turn repository content into structured findings such as dependency inventories, site maps, documentation plans, and other useful output. In the loop, skills answer a simple question: what does the project look like right now?
MCP is the handoff layer. It is how your agent moves those findings into gloria.dev and asks for the current picture back when it needs it. In the loop, MCP is the bridge that turns local analysis into shared state.
The web app is the shared record. It shows the current state of the project, keeps follow-up work visible, and sends notifications when something needs attention. In the loop, it is the part that keeps yesterday’s analysis from going stale.
Why the loop matters
A skill can read the source once, but the repository keeps changing. gloria.dev does not stop at analysis. Your agent records what it found, and the service keeps that picture current as new code lands.
That matters because the code is the source of truth for what was built, while the project’s intent is the source of truth for what was wanted. When those two drift apart, gloria.dev makes the gap visible so the next agent run starts from a fresher view.
Where to go next
Read Skills and MCP for the handoff in more detail, Overview for the broader product story, or Architecture for the system boundaries.