← Back to directory
P

Photoshop MCP

Community
An MCP server that lets you control Adobe Photoshop in plain language, with 116 tools.
Category
Other #71 of 230
Stars
★ 786 Very popular
Transport
stdio (local process)
Runtime
Node.js 18+
Credentials
No credential needed
License
MIT
Last commit
60FMRS · C

Photoshop MCP turns Photoshop into something you can talk to: describe what you want, and the AI plans the steps while Photoshop executes them, with each multi-step recipe counting as a single undo. It exposes 116 tools covering everyday design work such as background removal, portrait retouching, export, carousels, bulk watermarking, color grading, and generative fill, and it plugs into Cursor, Claude Code, Claude Desktop, and VS Code over stdio, with a built-in web chat UI as an alternative. As an unofficial community project under the MIT license, it suits designers who want AI to simplify repetitive image work, but not environments without Photoshop or users with strict concerns about cloud AI and anonymous analytics; back up originals and stay mindful of batch operations and supply-chain risk.

Strongest · Setup experience 14/20 Weakest · Security and permissions 9/20

Reliability
10/20
Security and permissions
9/20
Maintenance
13/20
Documentation
14/20
Setup experience
14/20
Why each score
Reliability 10/20
Static evidence shows a coherent structure: the README describes 116 tools (100 atomic + 16 recipes), server.json declares the npm package @alisaitteke/photoshop-mcp at 1.7.15 over stdio, and the manifest version matches the package version with no visible contradictions. Docs reference structured errors and a troubleshooting guide, implying designed error paths. However, this review is source-reading only and nothing was executed, and the supplied material contains no CI workflow and no committed tests covering the MCP init handshake or the match between declared tools and real behavior (docs/development.md is cited but not provided). Runtime depends on Adobe Photoshop 2012+ and Node.js 18+, both external and uncontrollable, and behavior without Photoshop is undisclosed. Per the static calibration cap of 12, and with key paths not reproducible, the score is held at 10.
Security and permissions 9/20
No red line is crossed: no malware, no credential theft, no covert exfiltration. The README explicitly discloses default anonymous aggregate usage analytics that can be disabled at any time, with a dedicated doc, which counts as data-flow disclosure. The unofficial, community-maintained identity is stated plainly with no fabricated maintainer. Install examples contain no real tokens or secrets. Deductions: this server can drive local Photoshop to perform deletes, bulk rewrites and exports, yet the material never states whether side-effecting operations require per-action confirmation or scoping. The standalone UI path asks the user for an AI provider API key (or reuse of Claude Code / Gemini CLI accounts), but key storage location, encryption and permission boundaries are not described in the provided material. No least-privilege or sandboxing statement is present. Given that, 9 is the conservative score.
Maintenance 13/20
Evidence supports active maintenance: MIT license (consistent between repository metadata and README badge), version has reached 1.7.15, GitHub Releases badge present, multi-language READMEs (zh/es/de/ja/tr), an official website, CONTRIBUTING.md, and a clearly named maintainer (Ali Sait Teke) with contact links; the npm name, repository URL and MCP Registry name are mutually consistent, indicating clear ownership. Only 1 open issue suggests healthy responsiveness, but no commit history, release cadence, dependency-update policy or security-response channel (e.g. SECURITY.md) is provided, and issue response time cannot be verified, so that is only a weak signal. The missing security channel and dependency-update path are the main deductions, yielding 13.
Documentation 14/20
Documentation is layered: install (three paths: standalone UI, Cursor plugin, generic mcp.json config), a tool list (docs/available-tools.md covering all 116 tools with parameters), standalone UI with auth modes, prompt layer and recipes, architecture, development, troubleshooting and an analytics note, plus multilingual coverage and concrete copy-pasteable commands. Limitations are disclosed: Photoshop must be running, Node 18+, generative fill needs an Adobe account, and the project is not affiliated with Adobe. Deductions: only the README and server.json were supplied, so the referenced docs/ files could not be checked for parameter accuracy, and cost/quota guidance is absent; version-compatibility scope and cross-platform differences are asserted mostly at badge level, i.e. hidden assumptions remain. These gaps keep it at 14 rather than full marks.
Setup experience 14/20
The install path is clear and short: a single npx -p @alisaitteke/photoshop-mcp photoshop-mcp-ui launches the standalone UI, and client integration is offered via Cursor and VS Code deep-link buttons, a one-line Claude Code command, and a generic mcp.json snippet, all copy-pasteable with no placeholder secrets. Prerequisites are explicit (Photoshop 2012+, Node.js 18+, Windows/macOS), and the npm version matches the manifest version, reducing version-mismatch risk. Deductions: a heavyweight external dependency (local Photoshop install and running instance) is required and connectivity cannot be verified locally; the Cursor Marketplace listing is not yet live, so users must manually copy .cursor-plugin/, mcp.json and assets/ into a local directory and reload, which is manual and fragile; the supplied material offers no verifiable post-connection artifact such as a test script or health check. Under the static cap of 15, the score is 14.

Static review · not runListed 2026-09-18

Read the FMRS scoring method →

Fit and risk

What it can accessRuns commands or code

Best for

  • Designers and creators who want to drive Photoshop conversationally without writing scripts or code
  • Teams with repetitive or batch image work (watermarking, multi-platform export variants)
  • Users already working inside MCP clients such as Cursor or Claude Code
  • Workflows that need reversible, single-undo multi-step image operations

Not for

  • Environments without Photoshop installed or running (the project depends on local Photoshop)
  • Enterprises expecting official Adobe support or endorsement (this is an unofficial community project)
  • Sensitive scenarios where prompts or document content must never be sent to an AI provider
  • Environments that cannot install Node.js 18+ or fetch software via npm/npx

Required permissions

  • Read and modify documents open in local Photoshop (layers, masks, color, exports, etc.)
  • Run a local Node.js process that communicates with the MCP client over stdio
  • Access specified local image folders for batch processing
  • Use an AI provider API key, or reuse Claude Code / Gemini CLI account credentials, in standalone UI mode
  • Collect anonymous aggregated usage analytics by default (can be disabled)

Risks and side effects

  • AI-planned steps modify Photoshop documents directly; although recipes are single-undo, batch or broad operations can still cause unwanted changes, so keep backups of originals.
  • Generative AI features (generative fill, Firefly, etc.) require an Adobe account and may incur extra cost or be subject to service terms.
  • Using a cloud AI provider means prompt text and document information read by the AI may leave your machine.
  • It relies on a third-party npm package fetched and executed via npx, which carries supply-chain risk; pin versions and verify the package source.
  • Anonymous usage analytics are collected by default and should be turned off if privacy matters to you.
  • Batch operations modify or overwrite files in target folders, and misconfiguration can affect many files at once.

Setup

Before you start

Runtime:Node.js 18+

  1. Make sure Photoshop is installed and running locally (Windows or macOS, version 2012+), and install Node.js 18+.
  2. Option 1 (standalone chat window): run npx -p @alisaitteke/photoshop-mcp photoshop-mcp-ui — a chat window opens in your browser; sign in with an AI provider API key or reuse your existing Claude Code / Gemini CLI account.
  3. Option 2 (inside your AI app): in Claude Code run claude mcp add photoshop -- npx -y @alisaitteke/photoshop-mcp; or add this to your MCP client config (Cursor, Claude Desktop, VS Code, …): {"mcpServers":{"photoshop":{"command":"npx","args":["-y","@alisaitteke/photoshop-mcp"]}}}.
  4. To get the Photoshop icon in Cursor's MCP list, manually copy the plugin files into the local plugins directory as described in the README and reload the window.
.cursor/mcp.json
{"mcpServers":{"photoshop":{"command":"npx","args":["-y","@alisaitteke/photoshop-mcp"]}}}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "photoshop": {
      "command": "npx",
      "args": [
        "-y",
        "@alisaitteke/photoshop-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add photoshop -- npx -y @alisaitteke/photoshop-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Check that it works

With Photoshop running and the client config `npx -y @alisaitteke/photoshop-mcp` in place, the client's MCP tool list should show the photoshop server's tools (116 tools including recipes); try a prompt like "Remove the background from this portrait" to confirm it drives Photoshop.

Troubleshooting

  1. If tools stop responding, first confirm Photoshop is running and is version 2012 or later.
  2. Check that Node.js 18+ is installed.
  3. Verify the MCP client config uses command npx with args -y @alisaitteke/photoshop-mcp, and restart the client if needed.
  4. Let the AI read the structured error and retry; common fixes are in docs/troubleshooting.md.
  5. For generative features, confirm a working Adobe account is signed in.
  6. To rule out network or proxy issues, verify npx can fetch @alisaitteke/photoshop-mcp.

Things to try

Once connected, you can ask your AI assistant things like:

  • Remove the background from this portrait — keep it editable with a mask.
  • Enhance this portrait — smooth the skin and fix the tones, medium intensity.
  • Prepare this design for web, then export Instagram and X post variants.
  • Split this wide banner into a 5-slide seamless Instagram carousel.

Use cases

Background removal and subject isolation with a clean, editable mask
Portrait retouching: skin smoothing, tone fixes, dodge & burn setup
Web and social export: sRGB, sharpened, correctly sized for Instagram, X, and more
Carousel creation: split one wide design into seamless, numbered slides
Bulk watermarking: a whole folder of photos in one go, originals untouched
Color grading, sky replacement, and generative fill (Adobe account required)

Supported clients

Cursor
Claude Code
Claude Desktop
VS Code

Listed from the project's documentation, not tested by this site.

Overview

Photoshop MCP is a community-maintained, open-source MCP server that lets AI assistants — such as Cursor, Claude, or its own built-in chat window — drive Adobe Photoshop through natural language. You describe what you want (for example, "remove the background from this portrait and keep it editable with a mask"), and the AI first checks the document state, then plans and executes the steps. Every multi-step "recipe" lands as a single undoable step in Photoshop. The project states it ships 116 tools (100 atomic tools plus 16 one-step recipes) covering background removal, portrait retouching (skin smoothing, tone fixes, dodge & burn), web and social export (sRGB, resizing, sharpening), carousel splitting, batch watermarking, color grading, sky replacement, and generative fill (Adobe account required). It requires Photoshop to be installed and running locally (Windows or macOS, version 2012+) plus Node.js 18+. It connects to MCP clients over stdio via the npm package @alisaitteke/photoshop-mcp, and can also launch a standalone web chat UI from the command line. Licensed under MIT, it is an unofficial project not affiliated with or endorsed by Adobe Inc., and it collects anonymous aggregated usage analytics by default, which can be disabled.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 670 · Tools 11 Compare with this →

Source revision ba36d8f2a28e Data synced 2026-11-02 Read the FMRS scoring method