# PixelLayer

> Open-source image processing for AI agents: an agent skill, eight local MCP tools, a native Rust engine and verified installation packages.

PixelLayer 0.5.1 helps agents remove backgrounds, convert PNG/WebP/JPEG, resize images and trace SVGs on the user's machine. Skill-guided setup obtains verified downloads, reuses the model cache, prepares one client configuration and guides MCP reconnection. The public installer targets Windows x64 and needs no Python, compiler, GitHub account or GitHub CLI.

Path-based MCP exchanges carry file paths and metadata instead of inline image payloads. The website's up-to-99% token reduction refers to that image-payload comparison; total conversation cost and host-side visual review depend on the host. Setup may download packages and a model; image operations run locally. The engine preserves originals and confines image operations to the prepared workspace. The model/runtime have their own notices, while the engine, MCP and skill use MIT.

## Start here

- [Product and downloads](https://pixellayer.site/): Local image MCP, skill and current distribution.
- [Quick start](https://pixellayer.site/docs/quick-start.md): Load the skill, authorize complete setup, reconnect and process actual files.
- [Installation](https://pixellayer.site/docs/installation.md): Verified downloads, requirements, package verification and shared model cache.
- [Agent skill](https://pixellayer.site/docs/skill.md): Delivery, supported defaults and what setup can automate.
- [Connect a client](https://pixellayer.site/docs/clients.md): Generate and apply the selected MCP draft, then reconnect.

## Image workflows and tools

- [Overview](https://pixellayer.site/docs/overview.md): How the skill, native engine, MCP and distribution fit together.
- [Workspace boundaries](https://pixellayer.site/docs/workspaces.md): Prepare the real task root; use input_directory "." for images in its root.
- [Workflow examples](https://pixellayer.site/docs/usage.md): Background removal, conversion, resize and combined cut plus SVG.
- [Eight-tool reference](https://pixellayer.site/docs/tools.md): prepare_workspace, get_result_info, remove_background, convert_format, resize_image, vectorize_image, list_models and get_model_info.
- [MCP and results](https://pixellayer.site/docs/protocol.md): Stdio transport, typed requests, paths, metadata and errors.

## Installation and operation

- [Model and runtime](https://pixellayer.site/docs/models.md): Standard BiRefNet General ONNX FP32 on CPU; separate 928 MiB verified model download.
- [Configuration](https://pixellayer.site/docs/configuration.md): Workspace, cache, runtime, threads, precision and idle TTL.
- [CLI reference](https://pixellayer.site/docs/cli.md): Setup, configuration drafts, diagnostics, models and JSON calls.
- [Troubleshooting](https://pixellayer.site/docs/troubleshooting.md): Downloads, certificates, reconnection, workspace errors and visual review.
- [Uninstall](https://pixellayer.site/docs/uninstall.md): Remove one verified program version while preserving models, settings, skill and image workspaces.
- [Releases and distribution](https://pixellayer.site/docs/releases.md): Versioned downloads, checksums, updates and retention.

## Reference

- [Architecture](https://pixellayer.site/docs/architecture.md): Agent, skill, local MCP server, native processors and filesystem.
- [Runtime and security](https://pixellayer.site/docs/security.md): Local file scope, verified downloads and process boundaries.
- [Performance and token reduction](https://pixellayer.site/docs/performance.md): Image-payload comparison, cold/warm CPU calls and resource limits.
- [Licenses and model provenance](https://pixellayer.site/docs/licenses.md): MIT product license and pinned model/runtime notices.
- [Migration](https://pixellayer.site/docs/migration.md): Replace legacy Python setup and removed tools with the native release.
- [FAQ](https://pixellayer.site/docs/faq.md): Setup capabilities, privacy, model requirements and supported distribution.

## Official downloads and source

- [Stable manifest](https://storage.googleapis.com/pixellayer-prod-downloads-website-prod-juan/channels/stable.json): Current approved version and download integrity information.
- [0.5.1 manifest](https://storage.googleapis.com/pixellayer-prod-downloads-website-prod-juan/releases/0.5.1/release.json): Exact package and skill URLs, sizes and SHA-256 values.
- [Public source](https://github.com/juanmh10/pixellayer): Engine, CLI, MCP, skill, packaging and license notices.
- [0.5.1 GitHub release](https://github.com/juanmh10/pixellayer/releases/tag/v0.5.1): Same-version alternative downloads.

## Optional

- [Complete documentation in text](https://pixellayer.site/llms-full.txt): All current product chapters in one response; use individual chapters when less context is needed.
- [Documentation in Markdown](https://pixellayer.site/docs.md): Complete Markdown version of /docs.
- [Human-readable documentation](https://pixellayer.site/docs): Styled documentation and chapter navigation.