Integration
Amp moves across the codebase. Windcraft keeps the design fixed.
Amp is built to act agentically across your repo, with Sourcegraph’s code intelligence behind it. That reach is the point — and the design risk: one guessed value spreads. Windcraft gives Amp a local MCP server to read tokens and contracts from instead of inventing them.
Why agentic reach amplifies drift
Amp plans and edits across many files in a run, informed by code intelligence about how the repo fits together. But that intelligence is about structure, not your design values — it does not know your brand hue or your radius scale. So a plausible-but-off color chosen in one edit propagates wherever the agent reaches. The more capable the agent, the faster a design slip travels.
How Windcraft plugs into Amp
Amp reads MCP servers from its settings. Register Windcraft and Amp can pull the exact tokens, the component contract, and the page pattern before it writes UI — reading your values rather than guessing them.
- Run npx windcraft init in your repo to generate the config.
- Add the Windcraft server to Amp’s MCP settings (amp.mcpServers).
- Amp calls get_design_tokens / get_component_contract / get_page_pattern as it works.
- Run npx windcraft check to flag any drift before you commit.
Amp MCP config (settings.json)
{
"amp.mcpServers": {
"windcraft": {
"command": "npx",
"args": ["windcraft", "mcp"]
}
}
}FAQ
- Does this work with Amp’s Sourcegraph context?
- Yes, and they are complementary. Sourcegraph’s code intelligence helps Amp understand how your repo fits together; Windcraft gives it the design values to generate with. One is about structure, the other about your design language.
- Where do I add the Windcraft server?
- In Amp’s MCP settings, under amp.mcpServers. Amp launches npx windcraft mcp on demand; there is no long-running process to manage.
- Does my source code leave my machine?
- No. The MCP server and analyzer run locally. Only metadata — tokens, contracts, rule violations — is synced to the cloud.