Adobe Opens Photoshop and Premiere to ChatGPT and Claude via Local MCP
Adobe is letting outside AI models control its creative apps through a new Model Context Protocol, turning chatbots into design agents.
Tools · Source: TechCrunch
What happened
Adobe just dropped a massive AI update at its MAX event in Miami. The headline feature is a local Model Context Protocol. This protocol lets third-party chatbots like ChatGPT, Claude, and Gemini directly interact with Adobe applications. Your favorite AI can now access features inside Illustrator, Photoshop, After Effects, and Premiere. Adobe vice president Deepa Subramaniam confirmed they are exposing numerous endpoints and will rely on user feedback to decide how much control to hand over.
The company is also pushing its own Firefly AI into new territory. A private beta tool called Builder lets users create UI prototypes, stepping directly onto Figma territory. Firefly is also getting a video assembly tool that stitches together clips, images, and sound effects. Adobe wants its internal AI agent to be proactive and always on, gathering context about your work to assist with quick actions like removing backgrounds or generating video thumbnails.
Core apps are getting heavy upgrades today. Photoshop adds prompt-based relighting and generative style sliders. Premiere gets an object removal feature for video and a new color grading system for professionals. Lightroom can now fix blinking eyes and remove blemishes. Adobe is also integrating new models from its Topaz Labs acquisition, specifically Wonder 4 for upscaling and ImageHDR for converting standard images to high dynamic range.
Key facts
- Model Context Protocol — Allows ChatGPT, Claude, and Gemini to use Adobe features locally
- Builder — A new Firefly prototyping tool in private beta competing with Figma
- Wonder 4 and ImageHDR — Two new image models from Adobe acquisition Topaz Labs
- Dylan Field — Figma co-founder who claims MCP benefits frontier labs more
Why it matters
The walled garden of creative software is cracking open. By releasing a local MCP, Adobe is turning its massive suite of tools into an API for frontier models. Builders can now create workflows where a language model acts as the brain, and Adobe software acts as the hands. You do not need to build a new image editor to build an AI design agent. You just need to orchestrate the tools creatives already use. This validates the agentic future where software operates software.
This shifts the competitive landscape for design software. Figma co-founder Dylan Field recently argued that MCP benefits frontier labs more than anyone else. He is right. By making its tools controllable by OpenAI and Anthropic, Adobe risks becoming a dumb pipe for creative execution. But it also locks users deeper into the Creative Cloud ecosystem. If Claude can edit your Premiere timeline, you have no reason to cancel your Adobe subscription to try a new AI video editor. Standalone AI design startups will struggle to compete with this distribution.
For builders
Build agents on top of Adobe
Adobe is exposing endpoints across its core apps via MCP. Founders can build specialized agents that automate tedious workflows like video clipping or batch editing. Creative agencies will pay for tools that cut their busywork in half.
Figma faces a new threat
Adobe Firefly Builder is a direct shot at Figma prototyping dominance. If you are building design tools, watch how Adobe bundles UI generation with its existing creative suite. Standalone prototyping tools will lose pricing power.
Leverage third-party model integrations
Adobe already supports models from Runway, Black Forest Labs, and Kling. Builders should note that Adobe is becoming an aggregator of the best open and closed models. Competing on raw model quality is a losing game when incumbents just bundle them all.
My take
Adobe is making a dangerous but necessary bet. By letting Claude and ChatGPT drive Photoshop, they are admitting that chat interfaces are the new default UI. They will either become the ultimate execution engine for AI, or they will get reduced to a backend utility. Either way, the busywork of design is dead. I build AI products every day, and seeing legacy software open up to local agents is exactly the shift we need.
Original reporting: TechCrunch. This is my rewrite and opinion.