codex-photoshop-psd-automation
Enables Codex to orchestrate Adobe Photoshop via MCP to reconstruct AI-generated artwork into editable, validated layered PSDs, with automated saving, reopening, and verification.
README
Codex Photoshop PSD Automation
Turn AI-generated artwork into editable, validated layered PSDs.
A successful API call is not delivery evidence — the PSD is saved, closed, reopened, and validated.
AI image generators are excellent at producing final visuals. Commercial delivery often needs more: editable text, replaceable assets, independent objects, structured groups, and a PSD that can actually be opened and reviewed.
This project uses Codex and Photoshop MCP to assemble, validate, reopen, and verify Reconstructed Layered PSDs from AI-generated or isolated design assets.
AI Artwork / Isolated Assets
↓
Codex
↓
Photoshop MCP
↓
Adobe Photoshop
↓
Reconstructed Layered PSD
↓
Save → Close → Reopen → Validate
The problem
Putting a flat PNG inside a PSD is not an editable source delivery.
Clients may need editable copy, Smart Objects, replaceable assets, visible groups, transparent elements, and a layered file that still works after it is saved and reopened. Manual reconstruction is slow and error-prone; trusting an in-memory automation result is not enough.
What this project does
Codex can orchestrate Photoshop to:
- create structured PSD documents and named groups;
- place isolated assets as independent Smart Objects;
- preserve transparent PNG and Alpha behavior;
- create editable TEXT layers with exact content, font, size, alignment, and RGB validation;
- validate transforms, opacity, effective visibility, required group content, and duplicate layer names;
- reserve non-overwriting output versions;
- save, close, reopen, and inspect the persisted PSD;
- independently decode the final preview and retain failure evidence on errors.
Why this is different
Why not just put a PNG inside a PSD?
Because that does not prove the source file is editable or deliverable.
This workflow verifies that required assets remain independent, text remains editable, Smart Objects survive reopening, Alpha survives reopening, transforms remain correct, required groups are effectively visible, and previews are decodable. The final artifact is validated from disk rather than trusted because an API returned success.
Reconstructed Layered PSD
This project does not claim to magically restore the unknown original layers of a flat image. Instead, it reconstructs an editable PSD from complete backgrounds, isolated major elements, editable text, and declared layout rules.
Final AI concept
↓
Generate or extract a complete background and isolated elements
↓
Reassemble in Photoshop with Smart Objects and editable text
↓
Validate the persisted layered PSD
Real Photoshop evidence
Validated on Adobe Photoshop 2026 / 27.8.0:
| Gate | Result |
|---|---|
| PSD save → close → authoritative reopen | PASS |
| Editable TEXT content, alignment, RGB, exact font, and size | PASS |
| Smart Object transforms, opacity, and effective visibility | PASS |
| Required group visibility and recursive content | PASS |
| Transparent PNG / Alpha preservation | PASS |
| Case C multi-transparent-PNG composition | PASS |
| Independent preview decoding | PASS |
| Full-resolution preview pixel round-trip | PASS |
| Photoshop state restoration | PASS |
| Unit tests | 221 / 221 PASS |
Status: PRODUCTION READY FOR REVIEWED DELIVERY.
That means the technical delivery workflow is verified for commercial review. New customer artwork still requires human visual approval; this is not autonomous aesthetic approval.
Architecture
┌──────────────┐
│ Codex │
└──────┬───────┘
│ MCP
▼
┌────────────────────┐
│ Photoshop MCP │
│ TypeScript / Node │
└────────┬───────────┘
│ COM / ExtendScript
▼
┌────────────────────┐
│ Adobe Photoshop │
└────────┬───────────┘
▼
┌────────────────────┐
│ Editable PSD │
│ + Preview + Report │
└────────────────────┘
Quick start
Windows with Adobe Photoshop is required for real Photoshop integration.
npm ci
npm run build
npm run test:unit
Then configure the MCP server for Codex or another MCP client. See Windows installation and Codex MCP setup.
Useful validation commands:
npm run test:delivery-mcp
npm run test:production-closure
test:production-closure is a real Photoshop test and creates local, gitignored disposable evidence.
Example delivery
background.png
main-subject.png
element-01.png
element-02.png
element-03.png
↓
01_Background
02_Main Subject
03_Elements
04_Text
05_Effects
06_Adjustments
↓
delivery-v01.psd
delivery-v01.png
delivery-v01-report.json
Who is this for?
- AI designers delivering editable source files
- automation engineers and creative-tooling developers
- freelancers who need reliable PSD handoff
- MCP and agent developers working with desktop creative software
- Photoshop automation enthusiasts
Use cases
- AI packaging artwork → editable PSD
- marketing artwork → structured source file
- AI illustration → separated production layers
- automated PSD QA and delivery validation
- creative delivery pipelines built around Codex and MCP
Limitations
- This is reconstructed layered delivery, not lossless recovery of an unknown original PSD.
- New customer artwork requires human visual review.
- Transactional editing of arbitrary existing customer PSDs (Case D) is not part of v1.0.
- Long-batch and large-file stress testing are outside the current verified matrix.
- Real Photoshop automation currently targets Windows.
Upstream attribution
Built on top of alisaitteke/photoshop-mcp, with additional production-oriented PSD delivery, extraction, validation, recovery, safety, and real-Photoshop verification workflows.
The upstream project provides the MCP server, Photoshop connection and execution foundation. This project adds the delivery-oriented workflow and evidence gates described above; it does not claim the entire Photoshop MCP was created from scratch.
See upstream audit, delivery specification, and production readiness.
License
MIT. See LICENSE. The upstream attribution is retained.
If you are building AI-assisted creative workflows, Photoshop automation, or MCP tooling, a ⭐ helps others discover the project.
Recommended Servers
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.