A comprehensive Claude Code plugin combining Adobe Firefly's AI image generation with E2B Agent Sandboxes for scalable agentic engineering.

This repository provides two powerful capabilities:

1. Adobe Firefly MCP Server - AI-powered image generation, manipulation, and creative workflows
2. Agent Sandboxes (E2B) - Isolated, scalable sandbox environments for parallel agent experiments

An MCP (Model Context Protocol) server that integrates Adobe Firefly's AI image generation capabilities with Claude Code.

• Image Generation: Generate images from text prompts with customizable dimensions and styles
• Generative Expand: Extend images beyond their original boundaries
• Generative Fill: Replace or fill portions of images using masks
• Background Removal: Remove backgrounds from images automatically
• Similar Image Generation: Create variations based on reference images
• Style Transfer: Apply artistic styles from reference images to new content

• Python 3.11+ with UV
• Adobe Developer Console account with Firefly API access
• Claude Code CLI

# Clone the repository
git clone https://github.com/nolandubeau/claude-code-adobe-firefly.git
cd claude-code-adobe-firefly

# Install and register the MCP server
cd apps/firefly_mcp
uv sync
uv run mcp install server.py --name "Adobe Firefly"

Set environment variables:

export FIREFLY_CLIENT_ID="your_client_id"
export FIREFLY_CLIENT_SECRET="your_client_secret"

1. Go to Adobe Developer Console
2. Create a new project or select an existing one
3. Add the "Firefly - Firefly Services" API to your project
4. Generate OAuth Server-to-Server credentials
5. Copy your Client ID and Client Secret to your .env file

The easiest way to register the MCP server:

cd apps/firefly_mcp
uv run mcp install server.py --name "Adobe Firefly"

Or manually add to ~/.claude/settings.json or .mcp.json:

{
  "mcpServers": {
    "adobe-firefly": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/apps/firefly_mcp", "python", "server.py"],
      "env": {
        "FIREFLY_CLIENT_ID": "your_client_id",
        "FIREFLY_CLIENT_SECRET": "your_client_secret"
      }
    }
  }
}

See CLAUDE.md for detailed tool parameters and usage.

Using Agent Sandboxes for complete agentic engineering control with isolation, scale, and full agency.

Watch the full video breakdown: Agent Sandboxes + Claude Code

Agent Sandboxes unlock 3 key capabilities:

• Isolation: Each agent fork runs in a fully isolated, gated E2B sandbox - safe from your local filesystem and production environment
• Scale: Run as many agent forks as you want, each independent with its own sandbox
• Agency: Agents have full control - install packages, modify files, run commands, handle more of the engineering process

SDK Features:

• Typer CLI with --use-mocks, --download, --show-images, --verbose
• Full API coverage: seed, aspect_ratio, style_options, structure
• 126 tests with comprehensive coverage

Create a .env file in the project root:

# Adobe Firefly (required for image generation)
FIREFLY_CLIENT_ID=your_firefly_client_id_here
FIREFLY_CLIENT_SECRET=your_firefly_client_secret_here

# E2B Sandboxes (required for sandbox operations)
E2B_API_KEY=your_e2b_api_key_here
ANTHROPIC_API_KEY=your_anthropic_api_key_here

# Optional: Required for git push/PR functionality
GITHUB_TOKEN=your_github_token_here

Get your API keys:

• Firefly Credentials: Adobe Developer Console - Create a project with Firefly Services API
• E2B API Key: https://e2b.dev/docs
• Anthropic API Key: https://console.anthropic.com/
• GitHub Token: GitHub PAT docs - Create with repo scope

Work from fundamentals to full workflows:

1. E2B Fundamentals - apps/sandbox_fundamentals/
2. CLI Tools - apps/sandbox_cli/
3. MCP Server - apps/sandbox_mcp/
4. Workflows - apps/sandbox_workflows/

cd apps/sandbox_fundamentals
uv sync

# Run examples in order
uv run python 01_basic_sandbox.py
uv run python 02_list_files.py
uv run python 03_file_operations.py
# ... through 13_expose_vite_vue_webserver.py

cd apps/sandbox_cli
uv sync

uv run python src/main.py --help
uv run python src/main.py init
uv run python src/main.py sandbox create --template agent-sandbox-dev-node22
uv run python src/main.py exec <sandbox-id> "ls -la"

# Copy sandbox MCP config
cp .mcp.json.sandbox .mcp.json
# Edit with your E2B API key

# Start Claude Code
claude

# Check MCP status
/mcp

# Use natural language
prompt: init a new sandbox
prompt: run ls -la in the sandbox

cp .mcp.json apps/sandbox_agent_working_dir/.mcp.json
cp .env apps/sandbox_agent_working_dir/.env

cd apps/sandbox_workflows
uv sync
uv run obox <repo-url> --branch <branch> --model <opus|sonnet|haiku> --prompt "your task" --forks 3

# Test Firefly MCP with MCP Inspector
cd apps/firefly_mcp
uv run mcp dev server.py

# Run Firefly MCP directly
uv run python server.py

# Run Firefly SDK CLI (with mocks for testing)
cd apps/firefly_sdk
uv sync
firefly generate "A sunset" --use-mocks --verbose

# Run Firefly SDK examples
cd apps/firefly_examples
uv run python 01_generate_image.py "A sunset over mountains"

# Run SDK tests (99 tests)
cd apps/firefly_sdk
uv run pytest --cov=firefly_sdk

# Run MCP tests (27 tests)
cd apps/firefly_mcp
uv run pytest --cov

# Total: 126 tests

This plugin includes two specialized UX design skills for creating distinctive, production-grade interfaces:

Invoke skills via Claude Code:

Use the bencium-controlled-ux-designer skill to help me design this form

Use the bencium-innovative-ux-designer skill for this landing page

Controlled UX Designer:

• Always asks before making design decisions
• Prefers flat, minimal design without shadows
• Systematic approach to colors, typography, layouts
• Best for: Production apps, dashboards, forms

Innovative UX Designer:

• Commits boldly to distinctive aesthetic directions
• Uses shadows, gradients, textures when intentional
• Breaks from generic SaaS patterns
• Best for: Landing pages, marketing sites, portfolios

The docs/examples/ directory contains detailed integration patterns combining Firefly image generation with Bencium UX design skills:

1. App Design with Generated Hero Images

• Use bencium-innovative-ux-designer to design the page layout
• Use /firefly-generate to create hero imagery matching the design aesthetic
• Integrate generated images into the final design

2. Product Landing Page

• Use bencium-controlled-ux-designer for accessible, production-ready layouts
• Use firefly-workflow agent for product photography pipeline
• Generate background-removed product shots with custom backgrounds

3. Marketing Campaign Assets

• Design campaign layouts with bencium skills
• Generate on-brand imagery with Firefly style transfer
• Create size variations for social, web, and print

4. Parallel Design Brainstorming (NEW)

• Launch 4-10 sandbox forks with different design directions
• Each fork uses Bencium skills + Firefly for complete designs
• Compare all results side-by-side in hours instead of weeks

• Adobe Firefly Services Documentation
• Adobe Developer Console

• Claude Code Product Page
• Claude Code Documentation
• Model Context Protocol

• E2B Documentation
• Claude Agent SDK

MIT License - see LICENSE for details.