> One-line install of the modelux Skill so Claude Code, Cursor, and other agent harnesses automatically know your APIs, MCP tools, routing configs, and SDKs. # Install the Claude Code / Cursor skill The **modelux skill** is a single markdown file that teaches your AI assistant everything it needs to write code against modelux: the proxy API, the Management API, the MCP tool catalog, routing config selectors, both SDKs, and budget / decision-trace conventions. You don't have to install it to use it. Agents can fetch it directly from the URL below; install it as a Claude Code / Cursor skill only if you want it to autoload on relevant tasks. ## Just read the skill (no install) The canonical URL: ``` https://modelux.ai/skill/SKILL.md ``` Any agent that can fetch a URL (or be handed a URL by the user) can use modelux. Paste the URL into your prompt, or fetch the file once and drop it into context. It's plain markdown and always up to date. ## Install as a Claude Code / Cursor skill If you're using Claude Code, Claude Desktop, or Cursor, install the skill once and stop pasting context into every prompt — your assistant autoloads it when a task matches. One-liner: ```bash curl -fsSL https://modelux.ai/skill/install.sh | sh ``` This drops the skill at `~/.claude/skills/modelux/SKILL.md`. Restart your agent harness to pick it up. Override the install location with `SKILLS_DIR`: ```bash SKILLS_DIR=~/my-custom-skills curl -fsSL https://modelux.ai/skill/install.sh | sh ``` ### Manual install If you'd rather not pipe `curl` into `sh`, fetch the file yourself: ```bash mkdir -p ~/.claude/skills/modelux curl -fsSL https://modelux.ai/skill/SKILL.md -o ~/.claude/skills/modelux/SKILL.md ``` ## What it gives your assistant Once installed, your assistant gains built-in knowledge of: - The modelux proxy API (`api.modelux.ai/v1`) and its OpenAI-compatible request shape - The Management API at `app.modelux.ai/manage/v1` and where to fetch the OpenAPI spec - The modelux MCP server and how to manage your account from natural language - Both SDKs (`@modelux/sdk` for Node, `modelux` for Python) and their `X-Modelux-*` extensions - Routing config selectors (`@config-name`) and the supported policy types - Budget enforcement responses (`BudgetExceededError`, 402 status) - Decision traces and replay experiments The skill itself is intentionally short — it points your assistant at the canonical, always-up-to-date sources: - [`/llms-full.txt`](/llms-full.txt) — every docs page concatenated - [`/openapi.yaml`](/openapi.yaml) — the Management API spec - [`/docs.json`](/docs.json) — programmatic doc tree That way the skill never goes stale: when our docs change, the assistant sees the new docs the next time it fetches. ## Companion: connect the MCP server The skill teaches your assistant *what* modelux is. The [MCP server](/docs/guides/mcp-setup) lets it actually *do* things — create budgets, run experiments, rotate keys, query analytics. Install both for the full experience. ## Source The skill source lives in the public site repo and is served as a static asset. View it directly: [modelux.ai/skill/SKILL.md](/skill/SKILL.md).