@alisaitteke/photoshop-mcp
Advanced tools
MCP server for Adobe Photoshop — 89 tools (generative AI + recipes), standalone web UI. Control Photoshop from Cursor, Claude, or natural language.
Languages: English · 简体中文 · Español · Deutsch · 日本語 · Türkçe
v1.1+ — recipe workflows, fewer round-trips, snappier sessions. Standalone UI ships Action Plan (beta) for plan-then-execute runs.
Note: This is an unofficial, community-maintained project and is not affiliated with or endorsed by Adobe Inc.
A Model Context Protocol (MCP) server that enables AI assistants like Claude and Cursor to control Adobe Photoshop programmatically. This allows you to create designs, manipulate images, and automate Photoshop workflows through natural language commands while working in your IDE — or through the bundled standalone web UI, which supports both API keys and CLI subscription accounts (Claude Code / Gemini CLI). The UI also offers an opt-in Action Plan (beta) mode that plans every Photoshop step in one LLM call, then runs them in a single pass.
Designers and developers want to drive Photoshop from AI assistants, but raw ExtendScript calls are brittle: agents waste tokens on trial-and-error, layer types break filters, and one failed command leaves the document in an unknown state.
Photoshop MCP adds state awareness (get_state, get_preview, get_capabilities), recipe tools that wrap multi-step outcomes in a single undo step, and structured error envelopes so agents know what to try next. The optional standalone UI and Action Plan mode reduce round-trips for longer workflows — so natural language can actually ship pixels, not just suggest them.
Engineering deep-dive: docs/architecture.md.
Don't want to wire this into Claude Desktop or Cursor? The same package ships a fully local web UI that lets you chat with an AI model and drive Photoshop through this MCP server underneath. Connect with a provider API key or, for Anthropic and Google, reuse the OAuth session from Claude Code or Gemini CLI — no separate API key required.
npx -p @alisaitteke/photoshop-mcp photoshop-mcp-ui
That's it. A local server starts on 127.0.0.1 (random free port) and your
default browser opens the chat UI automatically.
Pick any of the following on first launch — use an API key or your existing CLI subscription account (Anthropic and Google):
| Provider | Models | API key | CLI account |
|---|---|---|---|
| Anthropic | Claude Sonnet / Opus / Haiku | console.anthropic.com | npm i -g @anthropic-ai/claude-code → claude auth login |
| OpenAI | GPT-5, GPT-4.1, o-series | platform.openai.com | — |
| Gemini 2.5 Pro / Flash / Flash-Lite | aistudio.google.com | npm i -g @google/gemini-cli → gemini auth login | |
| OpenRouter | 100+ models from any provider | openrouter.ai | — |
api_key (default) — Vercel AI SDK + your provider API key. Usage is billed
per token at API rates; the UI shows estimated cost per chat.cli_account — Uses your local Claude Code or Gemini CLI OAuth session.
No API key is stored; the UI probes claude auth status / gemini headless
to verify login. Usage counts against your subscription quota, not API
billing — the status bar shows "Included in subscription".You can switch auth method per provider in Settings without losing the other credential (e.g. keep an API key while trying CLI account, then switch back).
An optional execution mode in the standalone web UI for API key auth only
(cli_account always uses the default agentic flow). Turn it on with the
Action Plan toggle next to the model selector in the composer.
Instead of a per-step ReAct loop (model → tool → model → tool …), Action Plan:
The plan appears as a live todo list above the tool-call cards, with per-step
status (pending → running → done / error). Plans are persisted in chat
history so they survive reload. The toggle is off by default; the existing
agentic flow is unchanged when Action Plan is disabled.
Good for multi-step prompts such as "remove the background and export for web" where you want fewer model calls and faster end-to-end execution.
~/.photoshop-mcp/data.db (SQLite, chmod 600). API keys never leave your
machine; CLI mode inherits OAuth from ~/.claude/ or ~/.gemini/.Open Settings from the sidebar at any time:
| Action | API key mode | CLI account mode |
|---|---|---|
| Set up | Paste key → Save | Install CLI → auth login → Check connection |
| Switch away | Choose API key — stored key is kept | Choose Uses your account — key is not deleted |
| Custom binary | — | Optional CLI path if claude / gemini is not on PATH |
| Cost display | Per-token estimate in status bar | Included in subscription badge |
Auth method is stored per provider in ~/.photoshop-mcp/data.db (authMethod:
api_key or cli_account). Existing configs without authMethod default to
api_key and keep working unchanged.
photoshop-mcp-ui [--port 5174] [--host 127.0.0.1] [--no-open]
The UI server stores your provider API keys and can drive Photoshop, so /api/*
is not open to everything running on your machine. Each request must pass three
checks:
--host you bound to) on the
server's port. Blocks DNS rebinding.The browser never has to deal with the token: the server injects it into the
index.html it serves. For scripting, read it from
~/.photoshop-mcp/ui-session.json (chmod 600) and send it as x-psmcp-token or
Authorization: Bearer, or pin your own with PSMCP_UI_TOKEN before starting
the server. Requests without a valid token get 401 unauthorized.
stream-json (Google). All paths talk to this same
Photoshop MCP server over STDIO.claude auth login /
gemini auth login in Terminal).On top of atomic photoshop_* tools, the server ships an opinionated AI/prompt
layer that helps host LLMs (Cursor, Claude Desktop, etc.) translate vague user
requests into reliable Photoshop actions:
instructions — workflow contract advertised on MCP initialize
(ping once, state-before-action, prefer recipes, error recovery). See
src/prompts/instructions.ts.prompts primitive — 22 pre-engineered templates (15 recipe + 7 guide:
ps.enhance_portrait, ps.remove_background, ps.generative_fill, …)
via prompts/list and prompts/get.photoshop_recipe_* tools (remove
background, enhance portrait, prepare for web, export social variants, color
grade, frequency separation, batch mockup, organize layers, gradient fade,
sky blend, dodge & burn, remove distraction, split carousel, batch watermark,
passport photo). Each wraps steps in a single
Photoshop history state (one Undo reverts all). 89 tools total (74 atomic
photoshop_generative_fill, photoshop_generative_remove,
photoshop_generative_expand, photoshop_generative_upscale, photoshop_sky_replacement,
photoshop_generate_image (Firefly via ExtendScript; Adobe account + credits required).photoshop_neural_filter via optional UXP bridge plugin (uxp-plugin/):
skin smoothing, harmonize, depth blur, super zoom, and colorize (B&W → color).photoshop_get_state (cheap snapshot),
photoshop_get_preview (base64 JPEG for vision verification),
photoshop_get_capabilities (version-aware feature flags).code and
suggested_next_tool for self-correction.Full reference: docs/prompt-layer.md.
Verify parity: npm run verify:photoshop-prompts. Latest results:
docs/development.md#integration-test-results.
Below are example prompts you can use with AI assistants (Claude, Cursor, etc.)
when this MCP server is configured. Prefer recipe tools (photoshop_recipe_*)
for multi-step outcomes — each recipe is a single undo step. Use atomic
photoshop_* tools only for fine-grained edits no recipe covers.
Ping Photoshop and read capabilities for my installed version.
Get the current document state before changing anything.
Open portrait.jpg, get a downscaled preview so you can verify the subject.
After each major recipe, get another preview to confirm the result.
Every recipe wraps a multi-step outcome in a single undo step and maps 1:1 to a ps.* prompt template.
Remove the background from the active portrait layer.
Use Select Subject + a layer mask with a 2px feather. Keep the original pixels behind the mask.
The subject must be on the active layer — not a flat color fill.
Equivalent MCP prompt template: ps.remove_background with { feather_px: "2", keep_shadow: "false" }.
There's a tourist photobombing my landscape on the active layer.
I'll rough-select him with the lasso — then run the remove-distraction recipe with a 1px feather.
Content-aware fill only; don't touch anything outside the selection.
Equivalent MCP prompt template: ps.remove_distraction with { feather_px: "1" }.
Enhance the portrait on the active layer at medium intensity with skin smoothing.
Use the enhance-portrait recipe — I want frequency separation + auto-tone in one undoable step.
If the active layer is text or a Smart Object, rasterize first or pick a raster layer.
Show me a preview when done.
Equivalent MCP prompt template: ps.enhance_portrait with { intensity: "medium", skin_smoothing: "true" }.
Set up dodge & burn on the active portrait layer: a 50% gray layer in overlay blend mode.
I'll paint with a white/black brush myself — just prepare the non-destructive setup.
Equivalent MCP prompt template: ps.dodge_burn with { blend_mode: "overlay" }.
Set up frequency separation on the active raster layer with a 6px blur radius.
I will paint on the Low and High layers myself — do not apply extra smoothing.
Tell me which layers to edit when the stack is ready.
Equivalent MCP prompt template: ps.frequency_separation with { radius_px: "6" }.
Fade the isolated subject into the background from the bottom up.
Apply a bottom_to_top gradient on the layer mask, 0 to 100%. Keep the mask editable.
Equivalent MCP prompt template: ps.gradient_fade with { direction: "bottom_to_top" }.
The sky in my landscape is blown out. Blend ~/skies/sunset.jpg in as the new sky,
horizon at 45% of the frame height, feathered so the treeline stays natural.
Equivalent MCP prompt template: ps.sky_blend with { sky_image_path: "~/skies/sunset.jpg", horizon_pct: "45" }.
Apply a warm film color grade to the open document as non-destructive adjustment layers.
Use the apply-color-grade recipe with preset warm_film.
Preview the result when finished.
Equivalent MCP prompt template: ps.apply_color_grade with { preset: "warm_film" }.
Prepare the active document for web: sRGB, downscale, sharpen, export one optimized JPEG to ~/.photoshop-mcp/exports.
Then export Instagram post and X post variants as separate JPEGs from the same document.
List the output paths in a table.
Equivalent templates: ps.prepare_for_web, ps.export_social_variants.
I have a mockup PSD open with a Smart Object layer named "Screen".
Replace it with every PNG/JPG in ~/assets/mockups/ and export one JPEG per asset.
Do not place flat layers — swap the Smart Object so perspective is preserved.
Equivalent MCP prompt template: ps.batch_mockup_replace.
Organize the layer stack: rename by kind, auto-group related layers, preserve originals.
Run the organize-layers recipe, then list layers so I can review the new structure.
Equivalent MCP prompt template: ps.organize_layers.
Split the active document into a 5-slide seamless Instagram carousel.
Each slide should be 1080x1350 — crop the slices, don't letterbox them.
Give me the exported file paths in swipe order so I can upload them as-is.
Equivalent MCP prompt template: ps.split_carousel with { slides: "5", size: "1080x1350" }.
Watermark every photo in ~/photos/portfolio with the text "© Jane Doe 2026".
Bottom-right corner, 40% opacity, small margin from the edges.
Export watermarked JPEGs — never touch the originals.
Equivalent MCP prompt template: ps.batch_watermark with { assets_dir: "~/photos/portfolio", text: "© Jane Doe 2026", position: "bottom_right", opacity: "40" }.
Turn the open portrait into a US passport photo: white background, proper headroom,
exact 600x600 px at 300 DPI. Also give me a 10x15 cm print sheet with copies.
Note: framing is approximated from subject bounds — official acceptance is not guaranteed.
Equivalent MCP prompt template: ps.passport_photo with { spec: "us_2x2", make_sheet: "true" }.
Create a 1920x1080 Photoshop document with RGB color mode.
Add a light blue background layer and fill it with RGB(240, 248, 255).
Add centered text "Welcome" in 64pt font.
Save as welcome.psd to my Desktop.
Search Pexels for "mountain sunset" images.
Create a 1920x1080 Photoshop document.
Place the downloaded image and fit it to fill the entire canvas.
Apply a subtle Gaussian blur of 3px.
Increase brightness by 15 and contrast by 10.
Add white text "Adventure Awaits" centered at the top in 72pt.
Set the text opacity to 90% and blend mode to OVERLAY.
Save as adventure.jpg with quality 10.
Open photo.jpg from my Desktop in Photoshop.
Get state, then run the enhance-portrait recipe at low intensity.
If I only need quick tone fixes, apply auto levels, auto contrast, and unsharp mask (120%, 1.5, 0) on the active layer instead.
Adjust hue +15 and saturation +15, or use prepare-for-web when I'm ready to export.
For old black-and-white scans, run the colorize neural filter first (requires the UXP bridge plugin).
Save as enhanced-photo.jpg with quality 12.
Create a 1200x800 document.
Add a new layer named "Background" and fill with RGB(50, 50, 50).
Place logo.png at position (100, 100).
Fit the logo layer to 50% of its current size.
Set blend mode to SCREEN and opacity to 85%.
Add another layer, fill with RGB(255, 100, 50).
Set this layer's blend mode to MULTIPLY and opacity to 60%.
Merge all visible layers.
Save as composite.psd.
Create a 1080x1350 portrait document (Instagram story size).
Add a layer and fill with gradient-like color RGB(120, 40, 200).
Add text "SUMMER" at (540, 300) in 96pt.
Change text color to white RGB(255, 255, 255).
Set text alignment to CENTER.
Add another text "2026" at (540, 450) in 128pt, white color.
Apply Gaussian blur 2px to the background layer.
Save as summer-poster.png.
Open image1.jpg.
Resize to 1920x1080.
Apply auto contrast.
Apply subtle sharpen (amount 80%, radius 1.0).
Save as processed-1.jpg with quality 10.
Close without saving changes to original.
Repeat for image2.jpg and image3.jpg.
Create a 2000x2000 square document.
Place abstract-pattern.jpg and fit to fill document.
Duplicate the layer.
On the duplicate, apply motion blur at 45 degrees, radius 50px.
Set blend mode to OVERLAY and opacity to 70%.
Add centered text "MOTION" in 120pt white.
Apply a rectangular selection from (200, 200) to (1800, 1800).
Invert the selection and delete (to create a border effect).
Flatten the image.
Save as motion-art.jpg.
Create a 3000x2000 document at 300 DPI for print.
Place hero-image.jpg and fit to fill the canvas.
Duplicate the image layer.
On the duplicate, desaturate it completely.
Set blend mode to LUMINOSITY and opacity to 50%.
Create a new layer named "Overlay".
Fill with RGB(255, 150, 0) and set blend mode to SOFTLIGHT at 30% opacity.
Add text "PORTFOLIO" at top center (1500, 200) in 96pt.
Set text color to white.
Add subtext "2026 Collection" at (1500, 320) in 36pt.
Create a rectangular selection around the text area.
Create a layer mask on the overlay layer.
Merge visible layers.
Save as portfolio-cover.psd.
Export as portfolio-cover.jpg at quality 12.
Open my-photo.jpg.
Play the "Vintage Look" action from "My Actions" set.
Adjust brightness by -10 to darken slightly.
Save as vintage-photo.jpg.
Execute this custom ExtendScript code:
app.beep();
alert('Processing started!');
Apply Gaussian blur 15px to the active layer.
[Wait for result]
Actually, that's too much blur. Undo that.
Apply Gaussian blur 5px instead.
Or:
Get the history states to see what operations were performed.
Undo the last 3 operations.
Redo 1 step to bring back one operation.
If a recipe returns version_unsupported or generative_unavailable, call get_capabilities and tell me which Photoshop feature is missing.
If a tool fails with suggested_next_tool, follow that hint (e.g. rasterize_layer before a raster-only recipe).
Never guess — read get_state after a failure and propose the next single step.
photoshop-mcp-ui); API key or CLI
subscription auth per provider (Anthropic, Google)photoshop_* + 15 recipe photoshop_recipe_*No installation required! Just configure your MCP client:
npx @alisaitteke/photoshop-mcp
To hack on the repo locally, see From Source in the development guide.
Add to your Cursor settings (.cursor/config.json or workspace settings):
{
"mcpServers": {
"photoshop": {
"command": "npx",
"args": ["-y", "@alisaitteke/photoshop-mcp"],
"env": {
"LOG_LEVEL": "1"
}
}
}
}
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"photoshop": {
"command": "npx",
"args": ["-y", "@alisaitteke/photoshop-mcp"],
"env": {
"LOG_LEVEL": "1"
}
}
}
}
Can't find the config file? In Claude Desktop, open Settings → Developer → Edit Config — it opens the correct path for your install (the %APPDATA% path may differ on some setups).
Use the Claude Code CLI or the Claude Agent SDK with the same MCP server entry.
Recommended (CLI):
claude mcp add photoshop -- npx -y @alisaitteke/photoshop-mcp
Manual JSON — merge into your project .mcp.json or Claude Code MCP settings (see examples/claude-code-mcp.json):
{
"mcpServers": {
"photoshop": {
"command": "npx",
"args": ["-y", "@alisaitteke/photoshop-mcp"],
"env": {
"LOG_LEVEL": "1"
}
}
}
}
Agent SDK (TypeScript) — pass the same mcpServers block in your query() options:
import { query } from '@anthropic-ai/claude-agent-sdk';
const q = query({
prompt: 'List open documents in Photoshop',
options: {
mcpServers: {
photoshop: {
command: 'npx',
args: ['-y', '@alisaitteke/photoshop-mcp'],
env: { LOG_LEVEL: '1' },
},
},
allowedTools: ['mcp__photoshop__*'],
},
});
PHOTOSHOP_PATH: (Optional) Specify custom Photoshop installation pathLOG_LEVEL: Logging level (0=DEBUG, 1=INFO, 2=WARN, 3=ERROR)ANALYTICS_DISABLED: Set to 1 or true to disable anonymous usage analytics entirelyPOSTHOG_DISABLED: Legacy alias for ANALYTICS_DISABLEDANALYTICS_PROVIDER: Analytics backend — mixpanel (default) or posthog (rollback)MIXPANEL_TOKEN: (Optional) Override the Mixpanel project tokenMIXPANEL_API_HOST: (Optional) Mixpanel ingest host (default: https://api-eu.mixpanel.com)POSTHOG_KEY: (Optional, legacy) PostHog project key — used only when ANALYTICS_PROVIDER=posthogPOSTHOG_API_HOST: (Optional, legacy) PostHog ingest host (default: https://a.alisait.com)POSTHOG_UI_HOST: (Optional, legacy) PostHog UI host (default: https://eu.posthog.com)Full reference for all atomic photoshop_* tools (parameters, examples, and usage):
docs/available-tools.md.
Each tool returns comprehensive context information about the current state of Photoshop, including:
This allows AI assistants to maintain awareness of:
Example Response:
{
"applied": true,
"filter": "Gaussian Blur",
"radius": 10,
"wasRasterized": true,
"context": {
"hasDocument": true,
"document": {
"name": "design.psd",
"width": 1920,
"height": 1080,
"resolution": 72,
"colorMode": "RGBColorMode",
"layerCount": 3,
"hasSelection": false
},
"activeLayer": {
"name": "Background",
"kind": "NORMAL",
"opacity": 100,
"blendMode": "NORMAL",
"visible": true,
"locked": false,
"isBackground": false
}
}
}
This context helps AI assistants remember what document and layer they're working on across multiple commands.
claude auth login /
gemini auth login in Terminal; credentials live under ~/.claude/ and
~/.gemini/Important Note: While Photoshop 2022+ supports UXP for plugins, external automation via AppleScript/COM can only use ExtendScript. UXP is designed for internal plugins and cannot be invoked from external scripts. Therefore, this MCP server uses ExtendScript for maximum compatibility across all Photoshop versions.
Common connection, scripting, and logging issues:
docs/troubleshooting.md.
| Symptom | Likely cause | Fix |
|---|---|---|
cli_not_found | Claude Code / Gemini CLI not installed | npm i -g @anthropic-ai/claude-code or npm i -g @google/gemini-cli |
not_authenticated | No CLI OAuth session (API key / SDK auth does not count) | Run claude auth login or gemini auth login in Terminal, or switch to API key auth |
| SDK client works, UI CLI mode fails | SDK/API credentials are separate from Claude Code CLI OAuth | Use API key in the standalone UI, or log in with claude auth login for CLI account mode |
claude / gemini not on PATH | Custom install location | Settings → CLI path → Check connection |
| Chat works in IDE but not UI (CLI mode) | OAuth tokens are CLI-only | Use CLI account in UI; API keys and CLI sessions are separate |
| Gemini multi-turn feels forgetful | Headless CLI may start a fresh session each turn | Known limitation; history is prepended to the prompt (MVP) |
From-source setup, build, lint, integration tests (with latest results), and usage examples:
docs/development.md.
System design, data flow, platform abstraction, and UI agent modes:
docs/architecture.md.
Sharing on LinkedIn or social? Use images/og-social.png and
docs/social-preview.md for OG setup and post copy.
Contributions are welcome! Please read CONTRIBUTING.md before opening a PR.
Ali Sait Teke — Full-Stack engineer & AI-era software architect (Python, Go, Node.js, React, Next.js, Vue).
This project started from a practical question: how do you make Photoshop reliably controllable by LLMs without fragile one-off scripts? It grew into an MCP server with 80 tools, a recipe/prompt layer for dependable multi-step workflows, and a local web UI so creative work does not require an IDE.
What this codebase demonstrates: TypeScript systems design, MCP protocol integration, cross-platform desktop automation (macOS AppleScript / Windows COM), structured error recovery for agentic loops, and a production-minded local-first UI (Vue 3 + Hono + SQLite).
MIT
Anonymous, aggregated usage events are collected by default to improve the
product. You can opt out at any time. Full details:
docs/anonymous-usage-analytics.md.
FAQs
MCP server for Adobe Photoshop — 89 tools (generative AI + recipes), standalone web UI. Control Photoshop from Cursor, Claude, or natural language.
The npm package @alisaitteke/photoshop-mcp receives a total of 1,032 weekly downloads. As such, @alisaitteke/photoshop-mcp popularity was classified as popular.
We found that @alisaitteke/photoshop-mcp demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Company News
Open source maintainers are under more pressure than ever. We're raising our open source program from the Team plan to the Business plan, free.
Security News
The supply chain control that delays freshly published gems now covers lockfile generation and gem vendoring in Ruby projects.
Security News
During a UK cyber test, a Mythos 5 agent used sockpuppets, social engineering, and prompt injection to try to get a maintainer to merge malware.
