rawtherapee-mcp-server

rawtherapee-mcp-server

Enables AI-assisted RAW photo development via RawTherapee CLI, with a visual feedback loop that allows the LLM to see and iteratively edit images.

Category
Visit Server

README

RawTherapee MCP Server

CI PyPI Python License: MIT

Cross-platform Model Context Protocol (MCP) server for AI-assisted RAW photo development via RawTherapee CLI. Provides 49 tools for profile generation, image processing, visual previews, batch operations, device presets, luminance-based local adjustments, lens correction, film simulation LUTs, profile inheritance, and metadata privacy.

What makes it unique: The LLM can see the photos it's editing. Preview tools return inline Base64 images via MCP's ImageContent protocol, creating a visual feedback loop where the AI analyzes the image, adjusts settings, previews the result, and iterates — just like a human editor.

MCP Client Compatibility

Not all MCP clients handle inline images the same way. The visual feedback loop requires a client that renders ImageContent from tool responses and a backing LLM with vision capabilities.

Client MCP Support Image Display Visual Workflow Status
Claude Desktop Full Yes Full Tested
Claude Code Full No (terminal) Partial Tested — images not rendered in terminal, but processing and text analysis work fully
Cursor Full Should work Should work Untested — MCP docs indicate ImageContent support
Windsurf Full Should work Should work Untested
Cline Partial Unknown Unknown Untested — community reports suggest ImageContent may not render (#1865)
Zed Full Unknown Unknown Untested

Minimum requirements for the full visual workflow:

  • MCP client renders ImageContent (type: "image", data: base64, mimeType: "image/jpeg") from tool responses
  • Backing LLM supports vision/image analysis (e.g. Claude with vision)
  • Tool response size accommodates ~150KB previews (most clients: 1MB limit)

Text-only clients: All 49 tools work without inline images. Preview tools return file paths instead. The LLM can still read EXIF metadata, histogram statistics, generate profiles, batch process, and use luminance presets — the visual feedback loop is the only feature that requires image support.

Prerequisites

  • Python 3.11+
  • RawTherapee 5.9+ with CLI component installed
  • An MCP-compatible client (see table above)

Verify the RawTherapee CLI is reachable before proceeding:

Windows (PowerShell):

& "C:\Program Files\RawTherapee\5.12\rawtherapee-cli.exe" --version

macOS:

/Applications/RawTherapee.app/Contents/MacOS/rawtherapee-cli --version

Linux:

rawtherapee-cli --version

Installation

Two supported methods - pick the one that matches your priorities:

Method Choose when you want
uvx Minimal setup, no persistent install, standard MCP community pattern
pip + venv Explicit version control, predictable updates, easier debugging, air-gapped environments

Option A - uvx

No persistent install. uvx downloads, caches, and runs the server on demand.

Important: always pin a specific version. The unpinned form ("rawtherapee-mcp-server") locks to the first resolved version and will not auto-upgrade on its own. See Updating for the full refresh procedure.

<details> <summary><strong>Windows</strong> (<code>%APPDATA%\Claude\claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server@1.0.4"],
      "env": {
        "RT_CLI_PATH": "C:\\Program Files\\RawTherapee\\5.12\\rawtherapee-cli.exe",
        "RT_OUTPUT_DIR": "C:\\Users\\YourName\\Pictures\\rawtherapee-output"
      }
    }
  }
}

</details>

<details> <summary><strong>macOS</strong> (<code>~/Library/Application Support/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server@1.0.4"],
      "env": {
        "RT_OUTPUT_DIR": "/Users/yourname/Pictures/rawtherapee-output"
      }
    }
  }
}

RT CLI is auto-detected at /Applications/RawTherapee.app/Contents/MacOS/rawtherapee-cli. Set RT_CLI_PATH if your installation differs.

</details>

<details> <summary><strong>Linux</strong> (<code>~/.config/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server@1.0.4"],
      "env": {
        "RT_OUTPUT_DIR": "/home/yourname/Pictures/rawtherapee-output"
      }
    }
  }
}

RT CLI is auto-detected at /usr/bin/rawtherapee-cli. Set RT_CLI_PATH if needed.

</details>

Option B - pip + venv

Explicit install with a dedicated virtual environment. You control exactly which version is running and updates are transparent.

1. Create the venv and install

Windows (PowerShell):

python -m venv "$env:USERPROFILE\.rawtherapee-mcp-env"
& "$env:USERPROFILE\.rawtherapee-mcp-env\Scripts\pip.exe" install rawtherapee-mcp-server

macOS / Linux:

python3 -m venv ~/.rawtherapee-mcp-env
~/.rawtherapee-mcp-env/bin/pip install rawtherapee-mcp-server

2. Configure Claude Desktop

Point Claude Desktop at the venv's entry-point script directly. No activation step is needed - the path is absolute.

<details> <summary><strong>Windows</strong> (<code>%APPDATA%\Claude\claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "C:\\Users\\YourName\\.rawtherapee-mcp-env\\Scripts\\rawtherapee-mcp-server.exe",
      "args": [],
      "env": {
        "RT_CLI_PATH": "C:\\Program Files\\RawTherapee\\5.12\\rawtherapee-cli.exe",
        "RT_OUTPUT_DIR": "C:\\Users\\YourName\\Pictures\\rawtherapee-output"
      }
    }
  }
}

Replace YourName with your actual Windows username.

</details>

<details> <summary><strong>macOS</strong> (<code>~/Library/Application Support/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "/Users/yourname/.rawtherapee-mcp-env/bin/rawtherapee-mcp-server",
      "args": [],
      "env": {
        "RT_OUTPUT_DIR": "/Users/yourname/Pictures/rawtherapee-output"
      }
    }
  }
}

</details>

<details> <summary><strong>Linux</strong> (<code>~/.config/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "/home/yourname/.rawtherapee-mcp-env/bin/rawtherapee-mcp-server",
      "args": [],
      "env": {
        "RT_OUTPUT_DIR": "/home/yourname/Pictures/rawtherapee-output"
      }
    }
  }
}

</details>

3. Verify the install

Windows:

& "$env:USERPROFILE\.rawtherapee-mcp-env\Scripts\python.exe" -c "import rawtherapee_mcp; print(rawtherapee_mcp.__version__)"

macOS / Linux:

~/.rawtherapee-mcp-env/bin/python -c "import rawtherapee_mcp; print(rawtherapee_mcp.__version__)"

Expected output: the installed version number (e.g. 1.0.4).

After editing claude_desktop_config.json, fully quit and restart Claude Desktop - closing the window alone leaves a background process running on both macOS and Windows.

Client Configuration

Claude Desktop

The RawTherapee CLI path is auto-detected on most systems. Set RT_CLI_PATH if auto-detection fails or RawTherapee is in a non-standard location.

<details> <summary><strong>Windows</strong> (<code>%APPDATA%\Claude\claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_CLI_PATH": "C:\\Program Files\\RawTherapee\\5.12\\rawtherapee-cli.exe",
        "RT_OUTPUT_DIR": "D:\\Photos\\Processed"
      }
    }
  }
}

</details>

<details> <summary><strong>macOS</strong> (<code>~/Library/Application Support/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_OUTPUT_DIR": "/Users/you/Pictures/Processed"
      }
    }
  }
}

RT CLI is auto-detected at /Applications/RawTherapee.app/Contents/MacOS/rawtherapee-cli.

</details>

<details> <summary><strong>Linux</strong> (<code>~/.config/Claude/claude_desktop_config.json</code>)</summary>

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_OUTPUT_DIR": "/home/you/Pictures/Processed"
      }
    }
  }
}

RT CLI is auto-detected at /usr/bin/rawtherapee-cli, /usr/local/bin/rawtherapee-cli, or /snap/bin/rawtherapee-cli.

</details>

<details> <summary><strong>Development (from source)</strong></summary>

Use uv to run from the cloned repository:

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uv",
      "args": ["--directory", "/path/to/rawtherapee-mcp-server", "run", "rawtherapee-mcp-server"],
      "env": {
        "RT_OUTPUT_DIR": "/home/you/Pictures/Processed"
      }
    }
  }
}

</details>

Claude Code

# Published package
claude mcp add rawtherapee -- uvx rawtherapee-mcp-server

# Development (from source)
claude mcp add rawtherapee -- uv --directory /path/to/rawtherapee-mcp-server run rawtherapee-mcp

Cursor

Add to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_CLI_PATH": "/usr/bin/rawtherapee-cli",
        "RT_OUTPUT_DIR": "/home/you/Pictures/Processed"
      }
    }
  }
}

Windsurf

Add to your Windsurf MCP configuration:

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_CLI_PATH": "/usr/bin/rawtherapee-cli",
        "RT_OUTPUT_DIR": "/home/you/Pictures/Processed"
      }
    }
  }
}

Cline (VS Code)

Add to your Cline MCP settings:

{
  "mcpServers": {
    "rawtherapee": {
      "command": "uvx",
      "args": ["rawtherapee-mcp-server"],
      "env": {
        "RT_CLI_PATH": "/usr/bin/rawtherapee-cli",
        "RT_OUTPUT_DIR": "/home/you/Pictures/Processed"
      }
    }
  }
}

Note: Cline may not render inline images from tool responses. All text-based tools work normally.

For pip + venv config examples for all clients, see docs/CLIENT_COMPATIBILITY.md.

Quick Start

After installation and client configuration, try this workflow:

  1. "Analyze this photo"analyze_image reads EXIF, computes histogram, generates a thumbnail, and returns processing recommendations
  2. "Make it warmer with more contrast"generate_pp3_profile creates a profile with warm white balance and contrast boost, preview_before_after shows the difference
  3. "The shadows are too dark"add_luminance_adjustment adds a shadow recovery spot, preview_with_adjustments shows the result
  4. "Export for my phone"process_raw with device_preset crops and processes at the right aspect ratio

Updating

Both installation methods require a two-step procedure. Upgrading the package alone is not enough - Claude Desktop caches the MCP server's tool list. Always follow the full sequence: update package → fully quit Claude Desktop → relaunch.

With uvx

uvx creates a per-package receipt that locks the resolved version. --refresh re-downloads the same locked version; it does not upgrade. To get a new release, clear the receipt and update the version pin:

Windows (PowerShell):

uv cache clean rawtherapee-mcp-server
# Update the version pin in %APPDATA%\Claude\claude_desktop_config.json
# e.g. change @1.0.4 to @1.0.5
Get-Process *claude* -ErrorAction SilentlyContinue | Stop-Process -Force
Start-Process "$env:LOCALAPPDATA\AnthropicClaude\Claude.exe"

macOS:

uv cache clean rawtherapee-mcp-server
# Update the version pin in ~/Library/Application Support/Claude/claude_desktop_config.json
osascript -e 'quit app "Claude"' && sleep 2 && open -a Claude

Linux:

uv cache clean rawtherapee-mcp-server
# Update the version pin in ~/.config/Claude/claude_desktop_config.json
pkill -f claude && sleep 2 && claude &

With pip + venv

Windows:

& "$env:USERPROFILE\.rawtherapee-mcp-env\Scripts\pip.exe" install --upgrade rawtherapee-mcp-server
Get-Process *claude* -ErrorAction SilentlyContinue | Stop-Process -Force
Start-Process "$env:LOCALAPPDATA\AnthropicClaude\Claude.exe"

macOS / Linux:

~/.rawtherapee-mcp-env/bin/pip install --upgrade rawtherapee-mcp-server
# macOS:
osascript -e 'quit app "Claude"' && sleep 2 && open -a Claude
# Linux:
pkill -f claude && sleep 2 && claude &

Verify the update worked

After relaunch, ask Claude:

"What is the current rawtherapee-mcp-server version and how many tools are registered?"

Compare against CHANGELOG.md. If the version string is correct but new tools are missing, see docs/TROUBLESHOOTING.md.

Check GitHub Releases for full changelogs.

Available Tools (49)

Discovery & Configuration (5)

Tool Description
check_rt_status Check RawTherapee installation, version, CLI path, and server configuration
list_templates List all available PP3 templates (built-in and custom)
list_device_presets List all device/format crop and resize presets
list_raw_files Scan a directory for supported RAW files
list_output_files List processed output files in the output directory

Metadata & Analysis (5)

Tool Description
read_exif Read EXIF metadata with structured processing recommendations
analyze_image All-in-one analysis: EXIF + histogram + thumbnail + recommendations
batch_analyze Analyze multiple images with EXIF, recommendations, and thumbnails
get_image_info Get dimensions, format, file size with optional inline thumbnail
get_histogram RGB histogram with per-channel statistics, clipping, and SVG visualization

Profile Management (8)

Tool Description
generate_pp3_profile Create a PP3 profile from base template + parameters + device preset
read_profile Display PP3 profile contents in human-readable format
adjust_profile Modify specific parameters in an existing profile
compare_profiles Diff two profiles with optional visual A/B comparison
save_template Save a profile as a reusable custom template
create_template_from_description Create a template stub from natural language description
delete_template Delete a custom template
interpolate_profiles Blend two profiles by linear interpolation

Preview & Visualization (7)

Tool Description
preview_raw Quick preview JPEG with optional inline image return
preview_before_after Side-by-side neutral vs. profile comparison
preview_exposure_bracket Multiple EV stops rendered for exposure comparison
preview_white_balance Multiple WB presets with Kelvin values
batch_preview Thumbnails for multiple RAW files
preview_luminance_mask Grayscale mask showing local adjustment coverage
preview_with_adjustments Preview with all Locallab spots active

Processing & Export (4)

Tool Description
process_raw Process a RAW file to JPEG/TIFF/PNG with inline thumbnail
apply_template Apply a template to process a RAW file with optional device preset
batch_process Process multiple RAW files with the same profile
export_multi_device Export one RAW optimized for multiple devices in one call

Crop & Device (3)

Tool Description
adjust_crop_position Reposition crop (left/center/right, top/center/bottom, or pixel offsets)
add_device_preset_tool Create a custom device preset
delete_device_preset Delete a custom device preset

Local Adjustments (5)

Tool Description
add_luminance_adjustment Add luminance-based local adjustment (shadows/midtones/highlights/custom)
list_local_adjustments List all Locallab spots in a profile
adjust_local_spot Modify an existing Locallab spot
remove_local_adjustment Remove a Locallab spot
apply_local_preset Apply a predefined local adjustment preset with intensity scaling

Lens Correction (2)

Tool Description
apply_lens_correction Apply Lensfun auto-detect or Adobe LCP correction to a PP3 profile
check_lens_support Query the Lensfun database for distortion/vignetting/TCA coverage

Film Simulation / LUT Support (4)

Tool Description
list_luts Scan RT_HALDCLUT_DIR for HaldCLUT film simulation LUTs, grouped by category
apply_lut Write a HaldCLUT film simulation into a PP3 profile with configurable strength
preview_lut Render an inline preview of a RAW file with a film simulation applied
preview_lut_comparison Render 2–5 LUT previews side-by-side for quick comparison

Profile Inheritance (3)

Tool Description
create_profile_variant Derive a child PP3 from a parent template with section-level overrides
list_profile_variants List all variants with override summaries, optionally filtered by parent
update_base_profile Modify a base template and propagate changes to all child variants

Metadata Privacy (3)

Tool Description
inspect_metadata Classify JPEG/TIFF EXIF into sensitive/technical/rights buckets with privacy recommendations
strip_metadata Losslessly remove GPS, serial numbers, software, and owner tags from a JPEG
set_metadata Write copyright, artist, description, and keywords into a JPEG

Built-in Resources

PP3 Templates (5)

Template Description
neutral Minimal processing, camera white balance, basic sharpening
warm_portrait Warm tones (5800K), gentle contrast, skin-friendly saturation
moody_cinematic Cool tones (5200K), lifted blacks, reduced saturation, film look
vivid_pet Warm tones (5600K), boosted saturation and vibrance, strong sharpening
bw_classic Black & white via channel mixer, high contrast, strong sharpening

Device Presets (19)

  • Mobile (7): Samsung Galaxy S26 Ultra, Galaxy S25 Ultra, iPhone 16 Pro Max, iPhone 16, Google Pixel 9 Pro, generic 9:16, generic 9:19.5
  • Desktop (5): 4K UHD (3840x2160), WQHD (2560x1440), Full HD (1920x1080), Ultrawide 21:9 (3440x1440), Dual 4K 32:9 (7680x2160)
  • Photo Formats (7): 3:2 (35mm), 4:3, 16:9, 1:1 (square), 5:4, 4:5 (Instagram portrait), 2:3 (portrait 35mm)

Custom presets can be added via add_device_preset_tool and persist across sessions.

Local Adjustment Presets (7)

Preset Description
shadow_recovery Brighten shadows without affecting highlights
highlight_protection Compress highlights to recover detail
split_tone_warm_cool Warm shadows, cool highlights
midtone_contrast Add contrast to midtones only
shadow_desaturation Desaturate shadow areas for a clean look
amoled_optimize High contrast and deep blacks for AMOLED displays
hdr_natural Natural HDR look with lifted shadows and compressed highlights

Presets accept an intensity parameter (50 = default, 25 = half, 100 = double).

Environment Variables

Variable Default Description
RT_CLI_PATH Auto-detect Path to rawtherapee-cli binary
RT_OUTPUT_DIR ~/Pictures/rawtherapee-mcp-output Default output directory
RT_PREVIEW_DIR OS temp dir Preview image directory
RT_CUSTOM_TEMPLATES_DIR ./custom_templates Custom PP3 templates directory
RT_PREVIEW_MAX_WIDTH 1200 Max preview width in pixels
RT_JPEG_QUALITY 95 Default JPEG quality (1-100)
RT_LOG_LEVEL WARNING Logging level (DEBUG, INFO, WARNING, ERROR)

See .env.example for a documented configuration template.

Known Limitations

  • RawTherapee CLI must be installed separately — it is not a Python package and cannot be installed via pip
  • stdio transport only — designed for local desktop use, no HTTP/SSE server mode
  • RT 5.12 crop+resize bug — RawTherapee 5.12 silently ignores Crop when Resize is also enabled in the same PP3 profile; the server applies a crop-only workaround automatically
  • 1MB MCP tool response limit — preview resolution is managed to stay within Claude Desktop's response size limit
  • No pixel-level editing — no retouching, object removal, frequency separation, or content-aware fill
  • No multi-image compositing — no HDR merge, panorama stitching, or focus stacking
  • Luminance masks are tonal only — no spatial or subject-aware masking
  • Integration tests require a RawTherapee installation and are skipped by default

Troubleshooting

RawTherapee not found:

  • Run check_rt_status to see detection details
  • Set RT_CLI_PATH to the full path of rawtherapee-cli:
    • Windows: C:\Program Files\RawTherapee\5.12\rawtherapee-cli.exe
    • macOS: /Applications/RawTherapee.app/Contents/MacOS/rawtherapee-cli
    • Linux: /usr/bin/rawtherapee-cli

Paths with spaces on Windows:

  • Always use the full path in RT_CLI_PATH, including quotes if your shell requires them
  • Environment variables in MCP client config JSON do not need extra escaping

Preview images too large:

  • Reduce RT_PREVIEW_MAX_WIDTH (default: 1200)
  • Preview tools auto-thumbnail to stay within the 1MB response limit

Server not responding:

  • Verify the entry in your MCP client config is correct
  • Restart your MCP client after config changes
  • Set RT_LOG_LEVEL=DEBUG to see detailed logs on stderr

New version installed but old tools still showing in Claude: See docs/TROUBLESHOOTING.md - Stale tool list after update. Short version: fully quit Claude Desktop (it runs as a background process), then relaunch.

Docker

docker build -t rawtherapee-mcp-server .
docker run -i --rm \
  -v /path/to/photos:/photos \
  -v /path/to/output:/output \
  -e RT_OUTPUT_DIR=/output \
  rawtherapee-mcp-server

Development

See docs/DEVELOPMENT.md for full setup instructions.

git clone https://github.com/lucamarien/rawtherapee-mcp-server
cd rawtherapee-mcp-server
pip install -e ".[dev]"
make validate    # lint + format + typecheck + test + security + audit

Testing

# Unit tests (mocked, no RawTherapee required)
pytest -v

# Integration tests (requires RawTherapee installation)
pytest -m integration -v

MCP Inspector

npx @modelcontextprotocol/inspector uv run rawtherapee-mcp

Contributing

See CONTRIBUTING.md for development guidelines, coding standards, and PR checklist.

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured