AgentBrush

Product Introduction

1. Definition: AgentBrush is a Model Context Protocol (MCP) server that integrates directly into AI-powered coding editors. It serves as a specialized tool for AI agents (like Claude Code, Cursor, Windsurf, Codex CLI, and Gemini CLI) to generate, edit, and manage on-brand visual assets through a standardized interface, all without requiring the user to leave their development environment.
2. Core Value Proposition: The product solves the critical problem of visual inconsistency in AI-generated projects. By feeding AgentBrush a brand's visual identity once, it acts as a persistent visual taste-maker for the coding agent. This ensures every generated product shot, icon, Open Graph (OG) card, and mascot adheres to the same brand family, resulting in cohesive and professionally designed outputs that land directly into the user's project repository. It empowers AI agents to design as well as they code.

Main Features

1. MCP Server Integration: AgentBrush operates as a lightweight MCP server, a standard protocol that allows large language models (LLMs) in coding editors to communicate with external tools. This integration enables the AI agent to call AgentBrush's image generation and editing capabilities directly from the terminal or chat interface, fitting seamlessly into the existing developer workflow.
2. Brand-Preserving Image Generation: The core technology involves a persistent brand identity engine. Users define brand assets and guidelines, which AgentBrush remembers. When generating new images—whether from a text prompt or via image-to-image translation using a provided brand asset—the system applies this consistent visual language to maintain brand cohesion across all assets.
3. On-Editor Mask & Iterative Editing: A key technical feature is the integrated mask editor. When an image generation is "almost right," the user's coding agent can invoke this editor. The user then paints over a specific region (a mask) that needs change, and AgentBrush regenerates only that masked area. This allows for precise, iterative refinement of visual assets without starting the entire generation process over, saving significant time and tokens.
4. Token-Based, Transparent Pricing: The product implements a usage model based on a monthly token budget. Different plan tiers (Starter, Pro, Power) offer varying token allocations and access to different quality presets (e.g., realistic, illustration, pixel art, isometric, logo). This structure is designed for simple, transparent pricing with no hidden fees or per-seat taxes, catering to different levels of production workload.

Problems Solved

1. Pain Point: Developers and technical creators often face a fragmented workflow when creating visual assets for their projects. They must exit their coding editor to use separate AI image generation tools (like Midjourney or DALL-E), manage inconsistent styles, and manually import assets. This process is inefficient and breaks the coding "flow state," leading to visual designs that may look "vibe-coded" rather than professionally cohesive.
2. Target Audience: The primary users are AI-augmented developers, indie hackers, solo founders, and small product teams (e.g., using Cursor or Claude Code). Secondary audiences include startup marketers needing quick, branded visuals and UI/UX designers collaborating with developers on projects where code and assets are tightly coupled.
3. Use Cases: Essential scenarios include generating on-brand marketing creatives and social assets at scale for a product launch, creating consistent UI icons and mascots for a new application, producing AI product photography mockups without a physical studio, developing 2D game assets (sprites, tilesets) with a unified art style, and designing professional Open Graph images to ensure every shared link looks polished and on-brand.

Unique Advantages

1. Differentiation: Unlike standalone AI image generators (e.g., Midjourney, DALL-E) or generic design tools, AgentBrush differentiates itself by being natively embedded in the developer's IDE via the MCP standard. It eliminates context switching and focuses specifically on brand consistency and iterative precision through its mask editor, which is less common in prompt-only generation tools. It also compares favorably to hiring designers or using stock assets for speed and integration.
2. Key Innovation: The key innovation is the persistent brand identity layer within an MCP server. This allows the AI coding agent to maintain a long-term memory of visual preferences and brand assets, enabling it to generate contextually appropriate and stylistically consistent images on demand. The integration of a region-specific mask editor for iterative refinement directly within the agent's command-line interface represents a significant step toward more sophisticated, human-in-the-loop visual creation in AI-assisted coding.

Frequently Asked Questions (FAQ)

1. How does AgentBrush work with coding tools like Cursor or Claude Code? AgentBrush installs as a Model Context Protocol (MCP) server. Your coding agent (Cursor, Claude Code, etc.) is configured to communicate with it. When you request an image, the agent formats the prompt and calls AgentBrush via MCP, receives the generated asset, and typically places the file directly into your project directory.
2. What file formats and quality levels does AgentBrush support? AgentBrush generates images in common formats suitable for web and development, with quality controlled by your subscription plan. Starter plans support low and medium quality, while Pro and Power plans unlock high-quality outputs. It offers multiple style presets including realistic, illustration, pixel art, isometric, and logo.
3. Is there an additional cost for teams or per user? No, AgentBrush features flat pricing with no seat tax. The pricing tiers (Starter, Pro, Power) are based on a monthly token budget, which dictates the amount of generation. Multiple team members can potentially share the same token budget, making it a cost-effective solution for small teams without hidden per-user fees.
4. Can I use my own brand logos or product images with AgentBrush? Yes, this is a core feature called Image-to-image with your assets. You can feed an existing brand asset (like a logo PNG) into your prompt, and the agent will use it to compose a full scene or new asset that maintains the same identity and brand coherence in a new context.