Your coding agent already reads your codebase. It knows which service owns billing, which folder holds the webhook handlers, and what calls what. Until now, getting that picture out of it meant asking for a description and drawing the diagram yourself.
Now it can show you. Connect Claude Code, Cursor, Codex, or any other MCP client to Lycana, and your agent can read your diagrams and change them on a Lycana canvas while you watch. It works with the same nodes, links, boundaries, and notes you do, and its changes wait for you before they land.
Your agent proposes. You decide.
How it works
There are three steps: Connect, Propose, Accept.
Connect. Add Lycana's MCP address to your agent, or install the Lycana CLI. Then approve the connection in Lycana. When you approve it, you choose whether that connection proposes its changes or applies them.
Propose. Ask your agent to map a system, add a service, or regroup a boundary. Its changes appear on the canvas as a proposal. While it works, the canvas toolbar names it, such as "Claude Code is editing". A waiting proposal appears in the same place, with its number of changes and Accept and Reject beside them.
Accept. Accept or reject in one click. Accepted changes join your history, so you can undo them like any change you made yourself. Agents can't undo.
A few details make the review practical. When a proposal arrives, the canvas brings its changes into view, and you can open a list of them and jump to each one. Click an empty part of the canvas to see the diagram as it was, without the proposal. Click the number of changes to bring it back. Accept and Reject work in either view.
Proposals are the default, not a requirement. Once you trust a connection, you can let it apply its changes directly, from your account settings or from the Connected agents menu on the canvas. An agent can suggest this, but it can't change the setting itself. Some changes stay proposals either way: removing more than three objects, or clearing the canvas, always waits for you.
Map a codebase down to the file
The most useful thing an agent can draw is the system it is already working in. A node can name the folder or file it describes, so a diagram can follow your code from services to packages to files, and say what each one does.
Subcanvases go as deep as you need. A node for a service can hold a subcanvas of its packages, and each package a subcanvas of its files. Your agent can find any object by its name, its description, or the file it describes, at any depth, and read an outline of the diagram's levels without opening every canvas.
Map the code again after a refactor and the diagram updates in place, without duplicates. Nodes the agent drew before are matched to the files they name and changed, rather than drawn a second time.
MCP or CLI
Both offer the same abilities. Use whichever your agent speaks.
If your agent supports MCP, add Lycana's address. In Claude Code, that takes one line:
claude mcp add --transport http lycana https://api.lycana.ai/mcp
Then type /mcp in Claude Code and choose lycana to sign in.
If your agent is more at home in a shell, install the CLI:
npm install -g lycana
lycana login
lycana outline <diagram>
lycana login shows a short code that you approve at lycana.ai/activate, on any device. lycana --help explains every command, and it is written so that an agent can work from it alone.
Cursor, Codex, and Claude on the web or desktop each take a minute to set up, and agent keys cover machines that can't open a browser. The steps for each are on the Agents page.
What it costs
Agent access is free. On Free, agents work on your one saved diagram. Lycana Pro opens your whole library, so an agent can work across every diagram you keep. Agent changes don't count against your AI command or voice allowances.
In a team space, an agent can do only what you could. You still need to be a member with editing rights, and the organization's plan must be active. When your agent starts editing, teammates see you as editing, together with the agent's name. If a teammate is already editing the diagram, the agent's change is refused, and the agent is told who is editing.
Lycana doesn't run an agent for you. This is for the agents you already use, working on the diagrams you already keep.