Agents read your lenses before they write.
rkito provides an MCP server that loads your organisation’s design context into the agent before it writes code. Agents work precisely within design intent.
| Type | Context | State |
|---|---|---|
| Lenses | Active lenses for the domains being modified | Loaded |
| Pillars | Pillar definitions | Loaded |
| Ledger | Entries for referenced components | Loaded |
| CHI | Active Change Intentions | Loaded |
What the MCP loads into your agent.
Works with Claude Code today, or any MCP-compatible agent you connect by hand.
Active lenses for the domains being modified.
Pillar definitions.
Entries for the components the work references.
The scope other work already claims.
Where you can use rkito today.
Only Claude Code is built. The rest are on the way and are not yet installable.
Terminal and VS Code extension. A2D plugin and catalog.
A2D extension.
VS Code and CLI.
Desktop, CLI and cloud agents.
The @rkito/cli helper.
From a connection to governed code.
Install the Claude Code plugin from the marketplace, or add the rkito MCP server by hand from /settings/mcp.
The agent retrieves the approved design and stays within CHI scope.
It implements exactly what was decided. Nothing less, nothing more.
The same governance record applies at the PR, as with any other author.
With and without context.
Agents know the design rules before generating code, so almost nothing from agents is blocked.
Agents generate sound code that violates your specific architectural intent.