Design and Build with Anima

Overview

Anima is the design agent in your AI coding swarm. This skill gives agents design awareness and the ability to turn visual ideas into production-ready code.

There are two distinct paths depending on what you're trying to do:

Path A: Create & Publish (Full App Creation)

Build complete applications from scratch. No local codebase needed. Anima handles everything: design, code generation, scalable database, and hosting. You go from idea to live URL in minutes.

This path is powerful for parallel variant creation. Generate multiple versions of the same idea with different prompts, all at the same time. Pick the best one, then open the playground URL to keep refining. All without writing a line of code or managing infrastructure.

Create Anima Playgrounds by: Prompt, Clone URL, Figma URL

What you get:

• A fully working application in an Anima Playground
• The ability to generate multiple variants in parallel and compare them
• No tokens wasted on file scanning, dependency resolution, or build tooling
• Scalable database already connected
• Scalable hosting when you publish

Path B: Integrate into Codebase (Design-Aware Code Generation)

Pull design elements and experiences from Anima into your existing project. Use this when you have a codebase and want to implement specific components or pages from a Figma design url or an existing Anima Playground.

Flows: Figma URL to Code (codegen), Anima Playground to Code

What you get:

• Generated code from Anima playgrounds designs adapted to your stack
• Precise design tokens, assets, and implementation guidelines

Prerequisites

• Anima MCP server must be connected and accessible — OR use the CLI (npx @animaapp/cli)
• User must have an Anima account (free tier available)
• For Figma flows: Figma personal access token (stored via auth --figma-token or connected during Anima authentication)
• For headless environments: an Anima API token

Important: Timeouts

Anima's playground-create tool generates full applications from scratch. This takes time:

• p2c (prompt to code): Typically 3-7 minutes
• l2c (link to code): Typically 3-7 minutes
• f2c (Figma to code): Typically 2-5 minutes
• playground-publish: Typically 1-3 minutes

Always use a 10-minute timeout (600000ms) for playground-create, playground-publish, and CLI calls. Default timeouts will fail.

Setup

There are two ways to use Anima:

Option 1: MCP Server (recommended for MCP-capable agents)

Before attempting any Anima MCP call, verify the connection is already working. Try calling any Anima MCP tool. If it responds, you're connected. If it fails, the user needs to set up authentication. See the setup guide for details.

Option 2: CLI (recommended for shell-based agents)

No MCP setup needed. Just npx and a token.

IMPORTANT: Before running any CLI command, you MUST set up authentication first. Without auth, all commands will fail.

Step 1 — Check if auth is already configured:

npx @animaapp/cli auth --status

Step 2 — If not configured, ask the user for their Anima API token, then run:

npx @animaapp/cli auth --token <anima-token>

The token is stored at ~/.config/anima/credentials.json and persists across sessions. The user gets their token from https://dev.animaapp.com > Settings > API Keys.

Step 3 (only for Figma flows) — Store Figma token:

npx @animaapp/cli auth --figma-token <figma-token>

Alternative: Set environment variables instead of using the auth command:

export ANIMA_API_TOKEN=<token>
export FIGMA_TOKEN=<token>

Do NOT try to pass tokens inline, construct auth headers manually, or write credentials files yourself. Always use npx @animaapp/cli auth --token or environment variables.

Troubleshooting: If npx returns ENOTFOUND, ENETUNREACH, or other network errors, the sandbox/environment may not have network access enabled. Re-run the command with network-enabled permissions, then continue with the auth flow.

The CLI outputs JSON when piped (for AI tools) and pretty text in terminals. All commands support --json, --verbose, and --timeout.

Choosing the Right Path

Before diving into tools and parameters, decide which path fits the user's goal.

When to use Path A (Create & Publish)

• User wants to build something new from a description, reference site, or Figma design
• User wants a live URL they can share immediately
• No existing codebase to integrate into
• Goal is prototyping, exploring visual directions, or shipping a standalone app

When to use Path B (Integrate into Codebase)

• User has an existing project and wants to add a component or page from Figma
• User wants generated code files to drop into their repo, not a hosted app
• User already built something in an Anima Playground and wants to pull the code locally

Ambiguous cases

When still unclear, ask: "Do you want a live hosted app, or code files to add to your project?"

From Request to Prompt

Before calling any tool, the agent needs to decide: is this request ready to build, or does it need clarification? And if it's ready, how do you write a prompt that lets Anima shine?

When to ask vs when to build

Threshold rule: Can you write a prompt that includes purpose, audience, and 3-5 key features? Yes = build. No = ask.

Signals to just build:

• "Build a recipe sharing app where users can upload photos and rate each other's dishes" (clear purpose, audience implied, features named)
• "Clone stripe.com" (unambiguous)
• "Turn this into a live site" + Figma URL (clear intent and input)

Signals to ask:

• "Build me a website" (what kind? for whom?)
• "Make something for my business" (what does the business do?)
• "Create an app" (what should it do?)

When you ask, ask everything in one message. Don't drip-feed questions. If the user is vague and doesn't want to answer, skip clarification and use Explore Mode to generate 3 variants instead. Showing beats asking.

Crafting the prompt

Anima is a design-aware AI. Treat it like a creative collaborator, not a code compiler. Describe the feel of what you want, not the pixel-level implementation. Over-specifying with code and hex values overrides Anima's design intelligence and produces generic results.

Include in prompts: purpose, audience, mood/style, 3-5 key features, content tone.

Leave out of prompts: code snippets, CSS values, hex colors, pixel dimensions, font sizes, component library names (use the uiLibrary parameter instead), implementation details, file structure.

Bad (over-specified):

Create a dashboard. Use #1a1a2e background, #16213e sidebar at 280px width,
#0f3460 cards with 16px padding, border-radius 12px. Header height 64px with
a flex row, justify-between. Font: Inter 14px for body, 24px bold for headings.

Good (descriptive):

SaaS analytics dashboard for a B2B product team. Clean, minimal feel.
Sidebar navigation, KPI cards for key metrics, a usage trend chart, and a
recent activity feed. Professional but approachable. Think Linear meets Stripe.

Path A: Create & Publish

Step A1: Identify the Flow

Determine which flow to use based on what the user provides and what they want.

User has a text description or idea → p2c

The most flexible path. Anima designs everything from your description. Best for new apps, prototypes, and creative exploration.

User has a website URL → l2c

Use l2c to clone the site. Anima recreates the full site into an editable playground.

User has a Figma URL → f2c (Path A) or codegen (Path B)

Two sub-cases:

• "Turn this into a live app" or "Make this a working site" → f2c (Path A). Creates a full playground from the Figma design
• "Implement this in my project" or "Add this component to my codebase" → codegen (Path B). Generates code files for integration

Quick reference:

Step A2: Create

Prompt to Code (p2c)

Describe what you want in plain language. Anima designs and generates a complete playground with brand-aware visuals.

MCP:

playground-create(
  type: "p2c",
  prompt: "SaaS analytics dashboard for a B2B product team. Clean, minimal feel. Sidebar navigation, KPI cards for key metrics, a usage trend chart, and a recent activity feed. Professional but approachable.",
  framework: "react",
  styling: "tailwind",
  guidelines: "Dark mode, accessible contrast ratios"
)

CLI:

npx @animaapp/cli create -t p2c -p "SaaS analytics dashboard for a B2B product team. Clean, minimal feel. Sidebar navigation, KPI cards for key metrics, a usage trend chart, and a recent activity feed. Professional but approachable." --guidelines "Dark mode, accessible contrast ratios"

Only -t and -p are required. Defaults: --framework react, --styling tailwind, --language typescript.

Parameters specific to p2c:

Styling options: tailwind, css, inline_styles

Link to Code (l2c)

Provide a website URL. Anima recreates it as an editable playground with production-ready code.

MCP:

playground-create(
  type: "l2c",
  url: "https://stripe.com/payments",
  framework: "react",
  styling: "tailwind",
  language: "typescript",
  uiLibrary: "shadcn"
)

CLI:

npx @animaapp/cli create -t l2c -u https://stripe.com/payments --ui-library shadcn

Only -t and -u are required. Defaults: --framework react, --styling tailwind, --language typescript.

Parameters specific to l2c:

Styling options: tailwind, inline_styles

UI Library options: shadcn only

Language: Always typescript for l2c

Figma to Playground (f2c)

Provide a Figma URL. Anima implements the design into a full playground you can preview and iterate on.

URL format: https://figma.com/design/:fileKey/:fileName?node-id=1-2

Extract:

• File key: The segment after /design/ (e.g., kL9xQn2VwM8pYrTb4ZcHjF)
• Node ID: The node-id query parameter value, replacing - with : (e.g., 42-15 becomes 42:15)

MCP:

playground-create(
  type: "f2c",
  fileKey: "kL9xQn2VwM8pYrTb4ZcHjF",
  nodesId: ["42:15"],
  framework: "react",
  styling: "tailwind",
  language: "typescript",
  uiLibrary: "shadcn"
)

CLI (accepts full Figma URL — extracts file key and node IDs automatically):

npx @animaapp/cli create -t f2c --file-key "https://figma.com/design/kL9xQn2VwM8pYrTb4ZcHjF/My-File?node-id=42-15" --ui-library shadcn

Only -t and --file-key are required. The CLI parses Figma URLs and normalizes node IDs automatically. Defaults: --framework react, --styling tailwind, --language typescript.

Parameters specific to f2c:

Styling options: tailwind, plain_css, css_modules, inline_styles

UI Library options: mui, antd, shadcn, clean_react

Step A3: Publish

After creating a playground, deploy it to a live URL or publish as an npm package.

Publish as Web App

MCP:

playground-publish(
  sessionId: "abc123xyz",
  mode: "webapp"
)

CLI:

npx @animaapp/cli publish abc123xyz

The response includes the live URL for the published app.

Publish as Design System (npm package)

MCP:

playground-publish(
  sessionId: "abc123xyz",
  mode: "designSystem",
  packageName: "@myorg/design-system",
  packageVersion: "1.0.0"
)

CLI:

npx @animaapp/cli publish abc123xyz --mode designSystem --package-name @myorg/design-system

Explore Mode: Parallel Variants

This is Path A's secret weapon. When a user says "build me X" or "prototype X", generate multiple interpretations in parallel, publish all of them, and return live URLs for comparison.

Workflow:

1. Generate 3 prompt variants from the user's idea. Each takes a different creative angle:

   • Variant 1: Faithful, straightforward interpretation
   • Variant 2: A more creative or opinionated take
   • Variant 3: A different visual style or layout approach

2. Launch all 3 playground-create calls in parallel (one per variant, type p2c)

3. As each one completes, immediately call playground-publish (mode webapp)

4. Return all 3 live URLs so the user can pick a favorite or ask for refinements. Optionally, if you have a screenshot tool available, capture each page to show in the chat.

Timing: All 3 variants generate in parallel, so total wall time is roughly the same as one (~5-7 minutes creation + 1-3 minutes publishing). Expect results within ~10 minutes.

Tips for good variant prompts:

• Keep the core idea identical across all three
• Vary the visual approach: e.g., "minimal and clean", "bold and colorful", "enterprise and professional"
• Add specific guidelines to each variant to differentiate them
• If the user mentioned a reference site or style, incorporate it into one variant
• Follow the prompt crafting principles above: describe mood and purpose, not implementation details

Path B: Integrate into Codebase

Step B1: Identify the Flow

Step B2: Match Project Stack to Tool Parameters

Step B3: Generate Code

Figma to Code (direct implementation)

MCP:

codegen-figma_to_code(
  fileKey: "kL9xQn2VwM8pYrTb4ZcHjF",
  nodesId: ["42:15"],
  framework: "react",
  styling: "tailwind",
  language: "typescript",
  uiLibrary: "shadcn",
  assetsBaseUrl: "./assets"
)

CLI (writes files directly to disk):

npx @animaapp/cli codegen --file-key "https://figma.com/design/kL9xQn2VwM8pYrTb4ZcHjF/My-File?node-id=42-15" -o ./components --ui-library shadcn

Only --file-key is required. Defaults: --framework react, --styling tailwind, --language typescript, -o ./anima-codegen-output.

Use the response fields (snapshots, assets, guidelines) as design reference when implementing.

Download Playground to Local Files

MCP:

project-download_from_playground(sessionId: "abc123xyz")

CLI:

npx @animaapp/cli download https://dev.animaapp.com/chat/abc123xyz -o ./my-project

The CLI accepts playground URLs directly and extracts the session ID automatically.

MCP vs CLI Quick Reference

When to use MCP vs CLI:

• MCP: Agent has MCP support and Anima server is connected
• CLI: Agent can run shell commands, no MCP setup needed, headless environments

CLI defaults (all optional): --framework react, --styling tailwind, --language typescript (when react). Only the type flag and the type-specific input (prompt, url, or file-key) are required.

Additional References

• Setup guide
• MCP Tools Reference
• Examples
• Troubleshooting
• Anima MCP Documentation