dcc-mcp-photoshop
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@dcc-mcp-photoshoplist layers in the active Photoshop document"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
dcc-mcp-photoshop
Agent workflow
AI agents should use the shared gateway through dcc-mcp-cli; IDE users may
continue to use the MCP endpoint. Prefer typed skills and tools over raw scripts.
Install or update the CLI
dcc-mcp-cli is the preferred control path for every shell-capable agent. If
it is missing, ask the user before installing the latest official release:
# Linux/macOS
curl -fsSL https://raw.githubusercontent.com/dcc-mcp/dcc-mcp-core/main/scripts/install-cli.sh | sh
# Windows PowerShell
powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/dcc-mcp/dcc-mcp-core/main/scripts/install-cli.ps1 | iex"Keep an official build current through the release manifest:
dcc-mcp-cli update check
dcc-mcp-cli update applyupdate apply downloads and stages the latest CLI for the next launch. It
does not update a running dcc-mcp-server; update that server in its own
environment.
dcc-mcp-cli dcc-types
dcc-mcp-cli list
dcc-mcp-cli search --query "<task>" --dcc-type photoshop
dcc-mcp-cli describe <tool-slug>
dcc-mcp-cli call <tool-slug> --json '{"key":"value"}'dcc-types reports release-catalog support; list reports live sessions. If a
tool belongs to an inactive progressive skill, call dcc-mcp-cli load-skill <skill-name> --dcc-type photoshop before retrying. For post-task improvement,
attach a stable session id with --meta-json, query dcc-mcp-cli stats --range 24h --session-id <task-id>, then pass the bounded evidence to the
review_skill_improvement prompt from dcc-mcp-skills-creator.
Bring Adobe Photoshop to MCP-native AI agents.
dcc-mcp-photoshop turns Photoshop into a standards-compliant MCP Streamable HTTP backend via the adobepy Rust broker and UXP bridge. Agents can inspect documents, create and edit layers, apply text, filters, smart objects, selections, export images, and automate Photoshop workflows through typed tools instead of brittle ad-hoc scripts.
The Python MCP server communicates with Photoshop through the adobepy Rust broker (port 47391), which proxies between the Python SDK and the UXP bridge running inside Photoshop.
Related MCP server: Photoshop MCP Server
Why Use It
What you get | Why it matters |
40+ typed Photoshop tools across 8 bundled skill packages | Agents can call validated tools for document, layer, image, text, selection, filter, smart object, and script operations. |
adobepy Rust broker | High-performance local proxy between Python SDK and UXP bridge. |
Sidecar isolation | Python MCP server runs outside Photoshop's UI thread; adobepy broker handles all host communication. |
Gateway compatible | Works with |
Multi-channel distribution | Available as a PyPI package or standalone adapter binary; the host bridge ships with adobepy. |
Canonical lifecycle | One cross-platform CLI owns plan, staged install, receipt, repair, rollback, verify, and uninstall. |
Quick Start
Prerequisites
Follow the cross-platform Install SOP. It documents the pinned official adobepy CLI checksum release, env-only token contract, Adobe-signed Photoshop product attestation, and the currently blocked bounded UXP bootstrap.
1. Install dcc-mcp-photoshop
python -m pip install dcc-mcp-photoshop
dcc-mcp-photoshop install --json --dry-run
dcc-mcp-photoshop install --json --yes2. Configure your MCP client
Point your MCP client to the gateway URL. The gateway auto-discovers which DCC (Photoshop, Maya, etc.) to route each tool call to, so you only need one endpoint:
{
"mcpServers": {
"photoshop": {
"url": "http://127.0.0.1:9765/mcp"
}
}
}3. Smoke test
Search for Photoshop tools, load the photoshop-document skill, and list layers in the current document.Architecture
AI Agent (Claude Desktop / Cursor / Copilot)
│ MCP Streamable HTTP (gateway port 9765)
▼
dcc-mcp-server Gateway
│ auto-discovers DCC via capability index
▼
PhotoshopMcpServer [Python sidecar]
│ HTTP JSON-RPC (port 47391)
▼
adobepy Rust Broker [adobepy broker]
│ WebSocket (port 47391)
▼
adobepy UXP Bridge [bridges/uxp/photoshop/, TypeScript]
│ Photoshop UXP API
▼
Adobe Photoshop 2022+Key architectural decisions:
The adobepy Rust broker (port 47391) is the central hub — Python SDK and UXP bridge both connect to it.
Python MCP instances use OS-assigned ports; skill scripts use
adobe.photoshop.Photoshop()facade to call the broker.The adobepy UXP bridge (
bridges/uxp/photoshop/) handles all Photoshop DOM operations with 60+ typed methods.All Photoshop automation goes through the adobepy facade layer, which abstracts over the broker.
Tool Surface
Skills are lazy-loaded: only meta-tools are available initially. Use load_skill to expand:
search_skills / dcc_capability_manifest
-> load_skill("photoshop-document")
-> list_layers(...)photoshop-document
Document information, listing, and lifecycle management.
Tool | Description | Read-only |
| Get metadata about the active Photoshop document (name, size, resolution, color mode) | ✅ |
| List all currently open Photoshop documents with metadata | ✅ |
| Open an existing local image or Photoshop document and make it active | ❌ |
| List all layers in the active document. Set | ✅ |
| Close a Photoshop document by ID or the active document | ❌ |
photoshop-image
Document creation, export, canvas/image resize, flatten, and merge operations.
Tool | Description | Read-only |
| Create a new document with specified dimensions, resolution and color mode | ❌ |
| Export the active document to PNG, JPG, TIFF, or PSD | ❌ |
| Save the active document in its current format | ❌ |
| Resize canvas without scaling content | ❌ |
| Scale the image (resamples content) | ❌ |
| Flatten all layers into a single background layer | ❌ |
| Merge all visible layers | ❌ |
photoshop-layers
Full CRUD for layers plus visual property changes.
Tool | Description | Read-only |
| Create a pixel, group, or adjustment layer | ❌ |
| Delete a named layer | ❌ |
| Duplicate a named layer | ❌ |
| Rename a layer | ❌ |
| Set opacity (0-100) of a named layer | ❌ |
| Show or hide a named layer | ❌ |
| Set blend mode (normal, multiply, screen, overlay, etc., 27 modes) | ❌ |
| Fill a layer with solid color (hex) or transparent | ❌ |
photoshop-text
Text layer creation, editing, and inspection.
Tool | Description | Read-only |
| Create a new text layer with content and styling (font, size, color, alignment) | ❌ |
| Update text content or style of an existing text layer | ❌ |
| Get text content and style properties of a text layer | ✅ |
photoshop-selection
Selection creation, modification, and saving.
Tool | Description | Read-only |
| Get current selection bounds (top, left, bottom, right) | ✅ |
| Select the entire canvas | ❌ |
| Deselect any active selection | ❌ |
| Create a rectangular selection (top, left, bottom, right in pixels) | ❌ |
| Create an elliptical selection | ❌ |
| Invert the current selection | ❌ |
| Expand selection by N pixels | ❌ |
| Contract selection by N pixels | ❌ |
| Feather selection by N pixels | ❌ |
| Save selection as a named alpha channel | ❌ |
photoshop-filter
Apply filters to layers.
Tool | Description | Read-only |
| Apply Gaussian Blur (radius in pixels) | ❌ |
| Apply High Pass filter (radius in pixels) | ❌ |
| Apply Sharpen filter | ❌ |
| Apply Smart Blur (radius + threshold) | ❌ |
| Generic filter by method name with arguments | ❌ |
photoshop-smart-object
Smart Object operations for non-destructive editing.
Tool | Description | Read-only |
| Convert the active layer to a Smart Object | ❌ |
| Create a new Smart Object by duplicating the active layer | ❌ |
| Open the Smart Object for editing | ❌ |
| Replace Smart Object contents with an external file | ❌ |
photoshop-script
Execute named Photoshop Actions for advanced automation.
Tool | Description | Read-only |
| Execute a named Photoshop Action from an Action Set | ❌ |
photoshop-adjustment
Channel operations and low-level batchPlay access.
Tool | Description | Read-only |
| Execute raw ActionDescriptor batchPlay commands | ❌ |
| Get channel information for the active document | ✅ |
| Add a new alpha channel | ❌ |
| Remove a named channel | ❌ |
Runtime Features
Feature | Surface |
Capability manifest |
|
adobepy broker | Local Rust proxy (port 47391) with session management and timeout control. |
Persistent logging | Logs written to |
Lazy skill loading |
|
Canonical Installer
The dcc-mcp-photoshop CLI is the only lifecycle owner. The legacy
photoshop-setup skill and PowerShell script remain discoverable compatibility
routes, but they delegate operators to the same contract.
load_skill("photoshop-setup")Tools
Tool | Description |
| Check system prerequisites (Python, pip, Photoshop) |
| Install or upgrade dcc-mcp-photoshop via pip |
| Redirect legacy calls to the canonical install lifecycle |
| Start server in dev mode for testing |
| Verify bridge connection to Photoshop |
| Auto-configure MCP client configs for Claude Desktop, Cursor, VS Code |
See install.md for the standard verbs, flags, JSON schema, exit codes, three-platform host paths, repair behavior, receipt-only uninstall, and verify-to-usable acceptance gate.
Installation
Use the canonical Install SOP for host integration. Installing a wheel alone does not load the UXP bridge or prove Photoshop readiness.
PyPI
pip install dcc-mcp-photoshopOr via uvx:
uvx --with dcc-mcp-photoshop dcc-mcp-photoshop --gateway-port 9765From source:
git clone https://github.com/dcc-mcp/dcc-mcp-photoshop.git
cd dcc-mcp-photoshop
pip install -e ".[dev]"Standalone Binary
Each GitHub Release includes platform-specific binaries built with PyOxidizer. No Python runtime required.
Platform | Binary name |
Windows |
|
Linux |
|
macOS |
|
# Download the latest release binary (example: Linux)
curl -LO https://github.com/dcc-mcp/dcc-mcp-photoshop/releases/latest/download/dcc-mcp-photoshop-linux
chmod +x dcc-mcp-photoshop
./dcc-mcp-photoshop --helpadobepy UXP Bridge
The canonical lifecycle uses the approved adobepy CLI through env-only authentication, stages the generated bridge, commits a redacted receipt, and then requires a real Photoshop RPC. See install.md; direct bridge generation is not a second supported installer surface.
Configuration
MCP Client Configuration
Claude Desktop
{
"mcpServers": {
"photoshop": {
"url": "http://127.0.0.1:9765/mcp"
}
}
}The gateway URL (:9765) is a unified facade that aggregates tools from all registered DCCs. On tools/call, the gateway auto-discovers the correct DCC instance via the capability index — no need to specify the DCC type in the URL.
Cursor
In Cursor Settings → Features → MCP Servers:
Name: photoshop
Type: url
URL: http://127.0.0.1:9765/mcpVS Code (via MCP extension)
{
"mcp": {
"servers": {
"photoshop": {
"type": "http",
"url": "http://127.0.0.1:9765/mcp"
}
}
}
}Environment Variables
Variable | Description | Default |
| adobepy broker HTTP endpoint |
|
| Required broker authentication token | — |
| Broker target identifier |
|
| Optional fixed MCP instance port |
|
| Gateway competition port | — |
| Log directory |
|
| Log level (DEBUG/INFO/WARNING/ERROR) |
|
| Operation timeout in seconds |
|
| Shared FileRegistry directory for gateway discovery |
|
| Extra skill directories (colon-separated) | — |
| Global extra skill directories | — |
Bridge Protocol
The Python ↔ Photoshop communication goes through the adobepy Rust broker using the adobepy protocol (v0.2.0). The broker handles JSON-RPC 2.0 routing, session management, and timeout control.
For protocol details, see the adobepy protocol contract.
Skill Authoring
Skills use the adobepy facade layer:
from adobe.dcc_mcp import action_result
from adobe.photoshop import Photoshop
from dcc_mcp_core.skill import skill_entry
@skill_entry
def list_layers(**kwargs) -> dict:
"""List all layers in the active Photoshop document."""
app = Photoshop()
return action_result(
"Listed active Photoshop layers",
lambda: {"layers": [layer.name for layer in app.activeLayers]},
prompt="Use the layer names in the next Photoshop operation.",
)
def main(**kwargs):
return list_layers(**kwargs)
if __name__ == "__main__":
from dcc_mcp_core.skill import run_main
run_main(main)Skill directory structure
skills/
├── photoshop-document/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-image/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-layers/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-text/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-selection/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-filter/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-smart-object/
│ ├── SKILL.md
│ └── scripts/
├── photoshop-script/
│ ├── SKILL.md
│ └── scripts/
└── photoshop-adjustment/
├── SKILL.md
└── scripts/Setting custom skill paths
export DCC_MCP_PHOTOSHOP_SKILL_PATHS=/path/to/my/skills
dcc-mcp-photoshop --skill-paths /path/to/my/skillsCLI Reference
dcc-mcp-photoshop [OPTIONS]
Options:
--mcp-port PORT Optional fixed MCP instance port (default: OS-assigned)
--broker-url URL adobepy broker URL (default: http://127.0.0.1:47391)
--gateway-port PORT Gateway competition port
--server-name NAME Server name in MCP initialize (default: photoshop-mcp)
--skill-paths PATH Extra skill directories
--no-builtins Do not discover built-in skills
--verbose, -v Enable debug logging
--version Show version and exitPython API
import dcc_mcp_photoshop
handle = dcc_mcp_photoshop.start_server(broker_url="http://127.0.0.1:47391")
print(f"MCP URL: {handle.mcp_url()}")
# ... use with Claude Desktop, Cursor, etc.
handle.shutdown()Ports
Port | Service | Description |
47391 | adobepy broker | Rust broker HTTP/WS endpoint (Python SDK ↔ UXP bridge) |
OS-assigned | MCP instance | Direct Streamable HTTP endpoint discovered through CLI/gateway |
9765 | Gateway | Optional multi-DCC gateway proxy |
Gateway Mode (Recommended for Deployment)
This mode uses the standalone dcc-mcp-server for the MCP server with the adobepy broker as the Photoshop communication layer.
Terminal 1 — Start the adobepy broker:
adobepy brokerTerminal 2 — Start the MCP server:
dcc-mcp-photoshop --broker-url http://127.0.0.1:47391MCP clients connect to the gateway URL:
http://127.0.0.1:9765/mcp— Gateway proxy (recommended)dcc-mcp-cli list— discover direct instance URLs
UXP Bridge Setup
The UXP bridge template is part of adobepy, while the canonical lifecycle owns its safe generation and receipt:
dcc-mcp-photoshop install --json --yesThis stages receipt-owned bridge files only. The canonical agent lifecycle does
not automate UXP Developer Tool or treat an adobepy doctor call as installation
progress. It remains fail-closed until the bounded Photoshop bootstrap/load
contract in dcc-mcp/adobepy#67 is available, then continues with
dcc-mcp-photoshop verify --json.
For human-controlled development only, outside the Install SOP readiness claim:
Open Photoshop
Go to Plugins → Development → Load Plugin...
Navigate to adobepy's
bridges/uxp/photoshop/and selectmanifest.json
Bridge manifest
The bridge manifest is at bridges/uxp/photoshop/manifest.json in the adobepy repository:
Field | Value |
Plugin ID |
|
Name |
|
Host app | Photoshop (PS), min version 25.0.0 |
Network permissions | All domains (required for WebSocket to broker) |
Local file system | Request (user consent per operation) |
Development
git clone https://github.com/dcc-mcp/dcc-mcp-photoshop.git
cd dcc-mcp-photoshop
pip install -e ".[dev]"
pytest tests/Requirements
Photoshop: Adobe Photoshop 2022+ (UXP support required)
Python (pip path only): Python 3.8+
Dependencies (auto-installed with pip):
dcc-mcp-core >= 0.20.14, < 1.0.0adobepy >= 0.1.0websockets >= 12.0
Build (to build standalone binary from source): Python 3.8+, Rust toolchain, and PyOxidizer (
pip install pyoxidizer)
Path | Photoshop Required? | Python Required? |
pip install | Yes (UXP plugin) | Yes |
Standalone binary | Yes (UXP plugin) | No |
adobepy bridge template | Yes | No |
Version Compatibility
dcc-mcp-photoshop | dcc-mcp-core | adobepy | Sidecar Binary |
current main | >=0.20.14,<1.0.0 | >=0.1.0 | dcc-mcp-server >=0.20.14 |
0.1.0-0.1.26 | >=0.12.14,<1.0.0 | >=0.1.0 | dcc-mcp-server >=0.12.14 |
Distribution
Release artifacts per version:
dcc-mcp-photoshop-<version>-py3-none-any.whl— Python wheeldcc-mcp-photoshop-<version>.tar.gz— Python sdistdcc-mcp-photoshop-windows.exe— Windows binarydcc-mcp-photoshop-linux— Linux binarydcc-mcp-photoshop-macos— macOS binary
Roadmap
v0.1.0 — Foundation ✅
Package structure and API design
PhotoshopBridge WebSocket client scaffold
Skill authoring helpers
v0.2.0 — adobepy Integration ✅
adobepy Rust broker integration (port 47391)
Python SDK facade (adobe.photoshop)
40+ Photoshop skills across 8 packages
Selection, filter, smart object, script, adjustment tools
UXP code convergence (removed old bridge/uxp-plugin/)
Configuration externalization (config.py)
v0.3.0 — Skills & Polish (next)
Color adjustments (levels, curves, hue/saturation)
Layer styles and effects
Batch processing workflows
Performance optimizations
v1.0.0 — Production Ready
Photoshop 2025+ UXP API compatibility
Authentication and security hardening
E2E test coverage
Troubleshooting
Broker not reachable
adobepy broker not reachable at http://127.0.0.1:47391Start the broker:
adobepy brokerVerify the broker is running:
curl http://127.0.0.1:47391/v1/capabilitiesCheck the
ADOBEPY_BROKER_URLenvironment variable matches the broker endpoint
UXP bridge not connecting
Ensure Photoshop 2022+ (min version 23.0.0) is running
Run
dcc-mcp-photoshop status --json, then follow its single structured next stepCheck the bridge is loaded: Plugins → Adobe Python Bridge
No active document error
If skills return "No active document":
Open a document in Photoshop (File → New or File → Open)
Retry the skill call
Default ports used
Port | Service |
47391 | adobepy Rust broker |
OS-assigned | MCP instance |
9765 | Gateway competition port |
Contributing
Contributions are especially welcome from those with:
Adobe UXP / ExtendScript experience
Photoshop automation knowledge
WebSocket and JSON-RPC protocol experience
See CONTRIBUTING.md for guidelines.
License
MIT — see LICENSE.
This server cannot be deployed
Maintenance
Related MCP Connectors
Marketo MCP server for AI. 130 tools to operate Marketo from Claude, Cursor, or ChatGPT.
Real-time chat hub for AI agents — Claude Code, Cursor, Cline, Codex over MCP or REST.
- QuallaaOAuthcom.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA server that allows AI tools like Claude and Cursor to automate and control Photoshop through natural language commands, enabling tasks like editing PSDs, toggling layers, and generating marketing creatives.10MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI assistants to programmatically control Adobe Photoshop on Windows to create documents, manipulate layers, and manage image adjustments. It provides a bridge between the Model Context Protocol and the Photoshop Python API for automated graphic design workflows.302MIT
- AlicenseNot gradedqualityCmaintenanceEnables automation of Photoshop tasks such as PSD editing, layer toggling, text insertion, and exporting through MCP clients like Claude or Cursor.MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI assistants to control Adobe Photoshop programmatically, supporting recipe workflows, state awareness, and generative AI features through natural language commands.3,798434MIT