For scripts and agents
이 콘텐츠는 아직 번역되지 않았습니다.
JSON output
Section titled “JSON output”Every command accepts --json. When passed, the command writes a single JSON document to stdout with no other output and never prompts interactively (create behaves the same as when stdout is not a terminal).
xds list --jsonxds create my-app --name orbit --targets web-react --dry-run --jsonxds tokens color.text --jsonErrors are written to stderr as a single JSON document:
{ "error": { "kind": "error", "message": "There is no theme \"nope\". Themes: forest, indigo, ocean, plum, sunset" } }kind is error, usage (for invalid arguments), or cancelled. A failing check (doctor or contrast --min) is treated as a normal result rather than an error: it is written to stdout with "ok": false.
Exit codes
Section titled “Exit codes”| Code | Meaning |
|---|---|
| 0 | The command succeeded |
| 1 | The command failed, or a check did not pass |
| 2 | Invalid command-line arguments |
| 130 | An interactive prompt was cancelled with Esc or Ctrl-C |
AGENTS.md
Section titled “AGENTS.md”This repository includes an AGENTS.md file that documents the design system rules for coding agents, along with a CLAUDE.md file that imports it. xds create writes both files into each generated project, tailoring AGENTS.md to list only the clients included in that project.
Agent skill
Section titled “Agent skill”skills/xds provides a skill in the Agent Skills format. It shows coding agents how to use the CLI commands above and includes task-specific guides for scaffolding and themes, tokens, components, and verifying UI changes.
npx skills add kjmin-dev/xdsIn Claude Code, this repository can also be added directly as a plugin marketplace:
/plugin install xds --marketplace kjmin-dev/xdsMCP server
Section titled “MCP server”xds mcp exposes the CLI commands as MCP tools for agents that cannot run shell commands directly. It communicates over stdio and is built on rmcp, the official Rust SDK.
claude mcp add xds -- xds mcp{ "mcpServers": { "xds": { "command": "xds", "args": ["mcp"] } } }| Tool | Description |
|---|---|
list |
Lists the available targets and themes |
tokens |
Lists design tokens, optionally filtered by query and tier |
components |
Lists all components |
component |
Returns the details for a single component, including source code when code is true |
contrast |
Computes the contrast ratio between two colors or color tokens |
doctor |
Runs project health checks |
create, add, theme |
Scaffolds a project, adds targets, or applies a theme |
Each tool returns the same JSON payload as the corresponding CLI command with --json. create, add, and theme run in dry-run mode unless "write": true is passed. The read-only tools accept an optional project directory; when omitted, they inspect the server’s working directory, or fall back to the embedded template if the working directory is not a project.
Agents with shell access do not need the MCP server; running the CLI commands with --json produces identical output.