For GitHub Copilot

Copilot agent mode reads what the workspace gives it. Aestheria, the design context layer, is the design half nobody has been committing, so every engineer on the team generates against the same system.

Copilot scales your conventions, whichever ones it finds.

Copilot is the widest-deployed coding agent in most companies, which changes the shape of the problem. On a team of forty, the question is not whether one generated screen is on-system. It is that forty engineers are each prompting from a different mental model of the product, and the agent agrees with all of them in turn.

Copilot instruction files (.github/copilot-instructions.md) help and have the same ceiling every context file has: they can say "follow the design system" and cannot say which token a card border takes. The instruction scales perfectly. The ambiguity scales with it.

One file, committed, and the team shares a system.

VS Code and Copilot agent mode read the same MCP config, so connecting is one command or one file. Commit it and every clone of the repository has the design context on first open, with no per-developer setup to chase.

{
  "servers": {
    "aestheria": {
      "type": "http",
      "url": "https://mcp.aestheria.design/mcp"
    }
  }
}

Pick Agent in the Copilot Chat dropdown and the tools are there: the token list with the rule for each, components with their variants and states, page templates, and validate_snippet to check the result. The code --add-mcp one-liner does the same thing without editing a file.

What it changes at team scale.

A reviewer stops being the place the design system is enforced. The agent resolves the token instead of proposing one, reuses the component instead of writing a fourth variant, and can check its own output before anyone opens the pull request. The rules move from tribal knowledge into something every clone of the repository carries.

The same context reaches whatever else your team runs, because the layer is not a Copilot feature. An engineer in Cursor and an engineer in VS Code are reading one library.

• 405 tokens
• 6 themes
• 37 components
• 27 charts
• 63 templates

Questions

How do I connect Aestheria to Copilot?

Run the `code --add-mcp` one-liner from the VS Code docs pointed at `https://mcp.aestheria.design/mcp`, or commit a `.vscode/mcp.json` with the same server. Copilot agent mode reads that file, so the team is connected by cloning.

Does this replace copilot-instructions.md?

No. Keep it. The instructions carry your rules; Aestheria is the system those rules refer to, so the sentence "use our tokens" finally resolves to a list with the reasoning attached.

Does it work in Copilot chat, or only agent mode?

MCP tools are an agent-mode capability, so the live library needs Agent selected. The repo-file route works everywhere: drop the kit in the project and the tokens, components, and `AGENTS.md` are ordinary workspace context.

Can we pin a version for the whole org?

Yes. The downloadable kit is a version you commit and update on your own schedule, which is the usual choice where a security review has to see what changed. The hosted server always serves the current library.