adobepy

Shared Adobe desktop communication runtime for Python tools and DCC MCP adapters. Python talks to a local Rust broker; Adobe applications run thin UXP or CEP/ExtendScript bridges.

adobepy is intended to be the common Adobe host layer for projects such as dcc-mcp-photoshop and future Adobe adapters. The public Python surface mirrors Adobe's JavaScript DOM where possible, while adding Pythonic aliases for agent and script ergonomics.

Implemented pieces:

• Rust adobepy CLI with broker, doctor, install-bridge, and repl.
• python -m adobe helpers (doctor, install-bridge, broker) so a wheel-only install can be diagnosed and can stage a bridge from Python.
• Local JSON-RPC broker with per-session token, target, timeout, and capability gates.
• Authenticated, typed broker/host/bridge runtime identity attestation for exact-instance checks by DCC MCP adapters.
• Bounded Photoshop UXP bootstrap with one-time exact-instance binding and a typed verification continuation; no arbitrary plug-in, command, script, or UI fallback.
• Bounded Illustrator CEP bootstrap that binds the receipted bridge and selected product process to a one-time claim and fixed adapter verification command.
• Python SDK under adobe.core, adobe.raw, adobe.photoshop, adobe.indesign, adobe.premiere, adobe.after_effects, and adobe.illustrator.
• Structured, non-eval access to the complete installed InDesign, Premiere, After Effects, and Illustrator object models.
• Optional DCC MCP result helpers under adobe.dcc_mcp for adapter skill code.
• UXP bridge templates for Photoshop, InDesign, and Premiere that dispatch to host APIs and broker raw JavaScript escape hatches.
• CEP bridge templates for After Effects and Illustrator.
• IR validation, proxy-aware .pyi stub generation, and stub/runtime drift checks.
• vx just package distribution workflow.

Validate everything:

npm install
npm run test:all

Bridge bundles under bridges/**/dist are generated artifacts and are not tracked. npm run test:all builds them for local verification; release CI rebuilds them before packaging. Run npm run uxp:build and npm run cep:build before using install-bridge directly from a source checkout.

The distribution name is adobepy and the import name is adobe. The PyPI wheel carries the SDK and the python -m adobe entry point; the broker executable and bridge templates ship in the platform bundle on GitHub Releases. Run python -m adobe doctor to see which parts are present and where to obtain the missing ones.

Installation, standalone use, broker startup, bridge loading, and troubleshooting are documented in install.md.

The full test suite type-checks bridge code, builds and exercises UXP/CEP bundles with mocked host runtimes, validates IR contracts, checks Python coverage, and runs Rust workspace tests.

Python support starts at 3.8. The current SDK wheel is pure Python; any future native Python extension must publish cp38-abi3-* wheels rather than per-minor-version wheels.

Build a redistributable Windows package:

vx just package

The archive is written to dist/adobepy-0.1.0-windows-x64.zip with a sibling SHA256 file. Usage and packaging notes are in docs/usage.md and docs/distribution-contract.md. Architecture boundaries and Adobe API source tracking are documented in docs/architecture.md, docs/api-sources.md, and docs/dcc-mcp-integration.md.

Python facade example:

from adobe.photoshop import Photoshop

app = Photoshop()
for layer in app.activeLayers:
    print(layer.name)

doc = app.activeDocument
doc.selection.select_rectangle({"top": 10, "left": 10, "bottom": 256, "right": 256})
for channel in doc.channels:
    print(channel.name)

text = app.activeText
if text:
    text.set_contents("Hello from Python")
    text.character_style.update(size=36, tracking=20)
    text.paragraph_style.update(justification="center")

layer = app.activeLayer
if layer:
    layer.filters.apply_gaussian_blur(2)
    layer.smart_object.convert_to_smart_object()

doc.exports.jpg_high("C:/renders/demo.jpg")