Install Gen AI CLI And MCP

Install gen-ai CLI & MCP

Two ways to use gen-ai:

1. CLI — gen-ai <command> in the terminal. Best for scripting, CI, cron, piping.
2. MCP — @picsart/gen-ai-mcp exposes every capability to your agent (Claude Code, Codex, Cursor, Windsurf, ChatGPT) as native tool calls. Best when you want to stay inside your AI tool and ask in plain English.

You can install just the MCP if you never plan to use a terminal — the MCP server runs its own binary and doesn't require the CLI to be installed. Most users end up with both.

Auto-trigger

Activate this skill on any of these phrases (natural language):

• "install gen-ai", "install picsart cli"
• "set up the MCP", "add gen-ai to Claude Code / Codex / Cursor / VS Code / Windsurf / ChatGPT"
• "connect my agent to Picsart", "wire Picsart into my IDE"
• "first-run setup", "get started with gen-ai"
• gen-ai: command not found
• MCP tool not available
• Any time the user is about to run gen-ai generate or call gen-ai_generate but isn't authenticated yet

Decision: CLI, MCP, or both?

Step 1 — Install the CLI (optional if MCP-only)

Pick the platform the user is on. Default to the shell installer.

# macOS / Linux
curl -fsSL https://picsart.com/gen-ai-cli/install.sh | bash

# Windows (PowerShell)
iwr -useb https://picsart.com/gen-ai-cli/install.ps1 | iex

# npm (Node 22+ already present)
npm install -g @picsart/gen-ai

Verify:

gen-ai --version

If "command not found", restart the shell or source the profile the installer wrote to. On macOS/Linux, the installer adds ~/.gen-ai/bin to the PATH via ~/.zshrc / ~/.bashrc.

Step 2 — Authenticate

gen-ai login

Opens OAuth in the browser. After:

gen-ai whoami

Must return an email and a non-expired token. If expired, re-run gen-ai login.

For CI / headless: set the auth environment variables before running the CLI:

export PICSART_ACCESS_TOKEN="<token from picsart.com/developer-console>"
export PICSART_USER_ID="<picsart-user-id>"
gen-ai whoami

Step 3 — Install the MCP server

npm i -g @picsart/gen-ai-mcp

The MCP server is a standalone binary — it runs its own process and exposes tools to the agent. It uses the same auth as the CLI (via ~/.gen-ai/credentials or PICSART_TOKEN).

MCP-only install (no CLI needed)

If the user lives inside an AI IDE and never wants a terminal CLI:

npm i -g @picsart/gen-ai-mcp

Authenticate by setting PICSART_TOKEN in the agent's env config (see per-host instructions below). No gen-ai login needed.

Step 4 — Register with the agent the user is using

Claude Code / Codex

Both support claude mcp add and codex mcp add respectively. They share the same MCP protocol, so the server registration is identical:

# Claude Code
claude mcp add gen-ai-mcp -- gen-ai-mcp

# Codex (OpenAI CLI)
codex mcp add gen-ai-mcp -- gen-ai-mcp

Or ask the agent in chat:

"Add @picsart/gen-ai-mcp as an MCP server. Then run gen-ai_whoami to confirm auth."

Cursor — .cursor/mcp.json

{
  "mcpServers": {
    "gen-ai-mcp": {
      "command": "gen-ai-mcp",
      "env": { "PICSART_TOKEN": "..." }
    }
  }
}

VS Code Copilot — .vscode/mcp.json

{
  "servers": {
    "gen-ai-mcp": {
      "command": "gen-ai-mcp",
      "env": { "PICSART_TOKEN": "..." }
    }
  }
}

Windsurf — ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "gen-ai-mcp": {
      "command": "gen-ai-mcp",
      "env": { "PICSART_TOKEN": "..." }
    }
  }
}

ChatGPT (Agents / custom connectors)

Add as a custom connector pointing at the gen-ai-mcp binary (locally or via a hosted proxy). Pass PICSART_TOKEN via env.

Zapier / n8n / Make

All three support MCP servers as tool providers. Point them at the gen-ai-mcp binary (or a container that runs it) and pass PICSART_TOKEN via env.

Any MCP-compatible host

The server speaks standard MCP. It exposes these tools:

• gen-ai_generate — image / video / audio generation
• gen-ai_batch — manifest-driven batch runs
• gen-ai_extend — extend videos
• gen-ai_list_models — enumerate available models
• gen-ai_model_info — inspect a specific model (voices, params, cost)
• gen-ai_compare_models — render the same prompt across N models
• gen-ai_estimate — dry-run cost check before spend
• gen-ai_pricing — live per-call cost for a model
• gen-ai_history — list past generations (for audit / A/B)
• gen-ai_login / gen-ai_whoami — auth scoped to the user's account

Step 5 — Verify with a cheap generation

Use Recraft V4 (2 credits, ~$0.02):

gen-ai generate -m recraft-v4 -p "a single smoke test sphere" --ar 1:1 \
  --json --no-input | jq -r '.url' | xargs curl -L -o test.webp

Or, from inside the agent (natural language):

"Generate a 1:1 test image with recraft-v4 and save it to ./test.webp."

The agent should call gen-ai_generate natively. If it doesn't, re-check the MCP registration and restart the IDE/host.

Step 6 — Sensible defaults

gen-ai config set defaultModel recraft-v4      # cheap, fast default
gen-ai config set autoOpen true                 # open outputs automatically
gen-ai completion zsh >> ~/.zshrc               # tab completion
gen-ai config set downloadDir "$HOME/Downloads" # absolute local download dir

Step 7 — Set up brand guidance (optional but recommended)

If the repo has a brand.md with palette, typography, voice, and forbidden patterns:

Read it before generation and include the relevant constraints in prompts or prompt files. The current CLI does not have a default rules config or policy gate flag.

Troubleshooting

Fully headless setup (CI / Docker)

# Dockerfile
FROM node:22-alpine
RUN npm i -g @picsart/gen-ai @picsart/gen-ai-mcp
ENV PICSART_TOKEN=""
CMD ["gen-ai", "--help"]

# GitHub Actions
- run: npm i -g @picsart/gen-ai
- run: gen-ai whoami
  env:
    PICSART_ACCESS_TOKEN: ${{ secrets.PICSART_ACCESS_TOKEN }}
    PICSART_USER_ID: ${{ secrets.PICSART_USER_ID }}
- run: gen-ai whoami

For agents

• Before the first gen-ai generate call (or gen-ai_generate MCP tool) in a session, run gen-ai whoami. If it fails, stop and ask the user to run gen-ai login — agents cannot complete OAuth on their own.
• If the user is in an MCP-enabled host and the CLI isn't installed, don't insist on installing the CLI — use the MCP server directly.
• Never run gen-ai update mid-batch — it restarts the binary and kills in-flight jobs.
• Default model for verification: recraft-v4 — 2 credits is a cheap smoke test.
• After successful install, offer to use brand.md as prompt context if one exists in the repo.
• For CI / headless, set PICSART_ACCESS_TOKEN and PICSART_USER_ID, then run gen-ai whoami — never interactive OAuth.