RapidRAW + MCP 0.4.0

RapidRAW Fork 0.4.0

The application is version 0.4.0. The native MCP bridge and npm MCP host remain independently versioned; this release keeps the compatible npm host at @sheldonxxxx/rapidraw-mcp@0.2.0.

Release tag: fork-v0.4.0.

Highlights

• Manage AI repairs and local adjustments in one Local Edits workspace. Create, reorder, hide and remove edits, refine their selections, and copy supported selections between a repair and an adjustment without coupling their later changes. Shape Light and Surface Colour selections are adjustment-only.
• Send the active crop and matching selection to compatible AI Connectors. Generated pixels are mapped back through the photo's crop and orientation, while the surrounding original scene remains intact.
• In the Inpaint studio, the last result is applied automatically after a full batch completes. Stopped or failed batches keep completed candidates in history without applying one automatically. Candidate deletion now requires a second click to confirm.
• Reduce stale masks and repeated preview work with content-sensitive cache keys, bounded mask bitmap caching, and reusable Marigold surface components.

Generative editing requires a separately configured provider and model assets. Review connector data-transfer behavior before use and inspect generated edits at native size. Source-build and CI results do not establish fresh-machine acceptance of every packaged download.

Install

Download the package for your platform from the RapidRAW 0.4.0 release. The native MCP bridge is included. For the pinned npm host version, setup steps, and client configuration, see the agent setup guide.

RapidRAW MCP host v0.2.0

The independently versioned Node MCP host for RapidRAW fork release fork-v0.3.0.

Highlights

• Adds public retouch options for removal-mask expansion/feathering and preview-only effective-mask inspection.
• Documents the Qwen Remove workflow and current packaged MCP setup.
• Published as @sheldonxxxx/rapidraw-mcp@0.2.0 with npm Trusted Publishing and provenance.

Compatibility

• Fork application: fork-v0.3.0
• Native MCP bridge: 1.2.0
• Node.js: 22.12 or later

Install

npx -y @sheldonxxxx/rapidraw-mcp@0.2.0 --binary /absolute/path/to/rapidraw-mcp --workspace /absolute/path/to/workspace

The npm host does not include the native application or model weights. Install the matching native package from this repository's fork releases and follow AGENT_SETUP.md.

RapidRAW + MCP 0.3.0

RapidRAW fork release fork-v0.3.0 with the native MCP bridge and a more complete generative editing workflow.

Highlights

• Inpaint studio variations with persistent candidate history and up to four reference images for supported Qwen Image 2.1 and Klein workflows.
• Qwen Remove brush cleanup with source-aware colour matching and optional texture borrowing.
• Removal coverage expansion, outward feathering, protected areas, effective-mask preview, and saved regeneration settings.
• More responsive live previews, native pan/zoom transforms, mask overlays, and stale-response protection.
• Photographer-facing Shape Light, Surface Colour, and Depth Selection tools.
• MCP asset-descriptor round trips, Linux NVIDIA runtime packaging, and expanded denoise/provider workflows.

Compatibility

• npm MCP host: @sheldonxxxx/rapidraw-mcp@0.2.0
• Native MCP bridge: 1.2.0
• Node.js: 22.12 or later for the packaged MCP host

Installation

Download the package matching your operating system and architecture from the assets below, then follow AGENT_SETUP.md to connect the published MCP host. Upstream RapidRAW packages do not include this fork's MCP bridge.

Generative editing requires separately configured providers and model assets. Preserve original photographs and review native-detail results before relying on an edit. See the full release notes.

NVIDIA runtime v1.0.0 (Linux x86_64)

Optional NVIDIA CUDA user-space runtime for RapidRAW on Linux x86_64.

Contents: official ONNX Runtime 1.30.0 (CUDA 13 build), NVIDIA cuDNN 9.20.0 for CUDA 13, CUDA 13.2/13.3 user-space libraries (cuBLAS, cuRAND, CUDA runtime, NVRTC). Archive is about 1.1 GB.

Requires a compatible NVIDIA driver (supported baseline 595.58.03 or newer). No root, pip, or system CUDA toolkit needed. x86_64 only; Linux ARM stays on CPU.

Install: download the .tgz and .sha256 below, verify with sha256sum -c, extract under ${XDG_DATA_HOME:-$HOME/.local/share}/rapidraw/nvidia-runtime/, and point current at the versioned directory. Next launch activates automatically. See docs/mcp/onnx-cuda.md for details.

RapidRAW + MCP 0.2.0

RapidRAW + MCP 0.2.0

Give your AI agent a native RAW editor with reversible edits and rendered photo review. Release tag: fork-v0.2.0. This is the whole fork's version, independent of upstream RapidRAW. The native bridge reports 1.2.0; the Node MCP server package reports 0.1.1 (published separately via the mcp-v0.1.1 tag and npm Trusted Publishing).

Highlights

• Native Nonlocal Bayer RAW denoise backends: in-process ONNX Runtime (CPU or Linux CUDA, explicit provider, no fallback) and macOS-only direct CoreML.framework, with pinned installable bundles via install_model kind="nonlocal". NIND and BM3D remain available.
• Texture-only healing for clean repairs with mismatched fine texture.
• Independently versioned npm MCP host releases (mcp-vX.Y.Z) through a fail-closed GitHub Actions path with OIDC Trusted Publishing and provenance.
• MCP documentation reorganization under docs/mcp/ with historical evidence snapshots, a Codex client adapter, and a concise npm package README.

Installation

The fork installs as RapidRAW MCP with separate preferences and model storage; existing upstream settings are not automatically migrated. Download an asset matching your OS and architecture below. Standard packages include the native MCP bridge; assets marked tethering also enable camera tethering and require system libgphoto2 installed before launch. The packages are not developer-signed or notarized.

The desktop package alone does not connect an agent. Install Node.js 22.12+ and connect the pinned npm host @sheldonxxxx/rapidraw-mcp@0.1.1 per the agent setup guide. Model weights install separately through install_model.

See the fork changelog and report reproducible issues with OS, architecture, app version and redacted error text.

RapidRAW + MCP 0.1.0 Beta 1

Known Apple Silicon installation issue: this package has an incomplete app-bundle signature and macOS may report it as damaged. Repair instructions for a checksum-verified download are documented in the macOS Beta 1 signature troubleshooting guide. The published Beta 1 assets have not been replaced; the packaging fix is unreleased.

RapidRAW + MCP 0.1.0 Beta 1

Give your AI agent a native RAW editor with reversible edits and rendered photo review. This first fork beta brings together the desktop editor, 65 MCP tools, local mask and detail enhancement, and configurable AI Connector workflows.

Release tag: fork-v0.1.0-beta.1. This is the whole fork's version, independent of upstream RapidRAW. The native bridge reports 1.2.0; the Node MCP server package reports 0.1.0.

Highlights

• Develop RAW photographs, refine masks, compare alternatives, and export JPEG or high-precision 16-bit TIFF through the native engine.
• Keep MCP work recoverable with isolated working copies, saved recipes, undo/redo, named versions and portable sessions.
• Use optional local models for mask refinement, scene and portrait-part selection, experimental deblur and conservative 2× enlargement.
• Choose compatible AI Connector workflows, generation resolution and seed from the desktop panel or MCP. Generative editing requires a separately configured provider.
• Includes frontend type/lint fixes and corrections to preset strength, RAW rendering, export profiles and processing geometry. See the fork changelog.

Installation

The fork installs as RapidRAW MCP and uses its own application identifier for preferences, library state and downloaded models. Existing upstream settings and models are not automatically migrated. Photo sidecars remain beside their images and use the same .rrdata format.

Download an asset matching your operating system and architecture from this release. Standard packages include the native MCP bridge; assets marked tethering also enable camera tethering. Apple Silicon requires macOS 14+, and Intel Mac requires macOS 13+. The packages are not developer-signed or notarized.

Tethering variants require system libgphoto2 to be installed before launch. On macOS, use brew install libgphoto2; on Ubuntu, use sudo apt-get install libgphoto2-dev. The beta DEB installer does not install this optional dependency automatically. See the camera tethering setup for supported platforms and camera requirements. Standard packages omit this dependency.

The desktop package alone does not connect an agent. Install Node.js 22.12+, build the Node host from the matching tag and configure your MCP client using the beta package setup. Your client supplies the language model. Model weights and optional companions are installed separately.

Release validation

The downloaded standard Apple Silicon package passed a 67-call native MCP acceptance suite on an existing Mac, using Canon CR3 and DNG photographs. Checks covered reversible edits, JPEG and high-precision 16-bit TIFF exports, session recovery, and preservation of source images and sidecars. This is packaged-app acceptance on that host, not a fresh-machine installation test.

The tagged source passed frontend typecheck, lint, formatting, translation checks, 32 frontend tests, 121 MCP tests, and the native MCP CI checks. See the packaging run for each platform's build result. Native runtime acceptance of the Windows, Linux and Intel Mac downloads remains unverified.

Beta limits

Native MCP editing has been exercised on macOS with Metal and on a Debian 13/NVIDIA server over SSH with Xvfb. Those source-build results do not establish Windows MCP support or fresh-machine acceptance of every packaged download. Review each release's build results separately from photographic quality.

Fine hair and feather masks, panorama framing, genuine HDR/focus brackets and film-negative quality retain acceptance gaps. Deblur and enlargement are experimental. Test on working copies and inspect rendered results before relying on an edit. See the capability matrix for detailed limits.

Report reproducible fork issues through GitHub Issues, including your OS, architecture, app version and relevant error text. Remove private image paths and credentials from logs before sharing.

Based on upstream RapidRAW history through commit 40cfa3df9c039f9f6adfd77a2c759b1ae788a3cd. Thanks to Timon Käch and the upstream contributors. This fork retains the AGPL-3.0 license.