Your AI assistant can write a flawless explanation of how Adobe Experience Platform segments work. What it can’t do, out of the box, is build one. The model can reason about your Adobe environment all day — it just can’t reach in and act on it.
The Model Context Protocol (MCP) is what closes that gap. Over the last few months we built MCP servers for Adobe Experience Cloud and open-sourced three of them. Together they expose 73 typed tools that let Claude, Cursor, ChatGPT Desktop, and GitHub Copilot do real work inside your Adobe environment — not describe it, do it. All three are Apache 2.0 and install with a single command.
New to MCP? Start with What an Adobe MCP Server Actually Does — this post gets specific.
AEP MCP — the first full-featured server for Adobe Experience Platform
Adobe ships a read-only beta MCP for Journey Optimizer; ours is the first full-featured Model Context Protocol server for Adobe Experience Platform itself — 34 tools across 10 categories: schemas, datasets, identities, profiles, segments, sources, destinations, Query Service, Privacy Service, and Datastreams. Full read and write, working pagination, and OAuth 2.0 Server-to-Server auth against Adobe IMS with token caching and automatic re-authentication. In practice: ask your AI to create an XDM schema (aep_create_schema), run a query against the data lake (aep_run_query), or submit a GDPR/CCPA privacy job (aep_create_privacy_job) — each a single tool call. AEP is the foundation that Journey Optimizer, Customer Journey Analytics, and Real-Time CDP all sit on, so an MCP at this layer reaches far more than any product-specific one.
Firefly Services MCP — generative media plus the Photoshop and Lightroom APIs
19 tools across three Adobe APIs: 9 for Firefly (image generation, expand, fill, composite, video), 6 for the Photoshop API (smart objects, text editing, background removal), and 4 for the Lightroom API (tone, straightening, edits, presets). It’s a thin, honest adapter over Adobe’s official SDKs — “generate an on-brand image” (firefly_generate_image) or “expand this canvas from 1024 to 1408 px” (firefly_expand_image) runs as one call, with Adobe’s own APIs doing the work. All 19 tools are live-validated end-to-end against the Firefly Services sandbox.
EDS MCP — Edge Delivery Services from your editor
20 tools across publishing (preview, publish, unpublish, bulk operations, cache purge), content (page retrieval, search, metadata, sitemaps, redirects), and metrics (Core Web Vitals, 404 tracking, experiments, logs). So “preview and publish the homepage” (eds_preview_and_publish) or “find every pricing page missing a description” (eds_search_pages → eds_get_metadata) become things an agent does, not a person clicks through.
What “production-grade” means here
A lot of MCP servers are weekend demos. These are built for the senior engineers doing real work in enterprise Adobe environments, which sets a higher bar:
- They run locally as stdio processes — nothing routes through our infrastructure, no telemetry phones home, no rate limits we impose.
- Real OAuth, real write operations, working pagination, structured error codes — the unglamorous things that separate a tool you trust from one that breaks on the second call.
- All three are Apache 2.0 on npm and work with any MCP-compliant client, not just Claude. One command each (for example,
claude mcp add eds -- npx @focusgts/eds-mcp-server).
See all 73 tools, with install commands
Full tool tables, install instructions for Claude Code, Cursor, and any MCP client, and an honest side-by-side with Adobe’s own MCP releases — all on one page.
Explore the Adobe MCP servers →Why we’re the ones building this
Focus GTS works at the execution layer of Adobe Experience Cloud — we place senior specialists, run managed delivery through Navigator, and build the open-source tools the work runs on. These servers aren’t a side quest; the fastest way to learn where AI helps an Adobe team (and where it doesn’t) is to build the thing and put it in real hands. They’re also part of a wider open-source set we maintain — Claude Code skills for Firefly Services, content-ops skills for EDS, an autonomous EDS operations agent, and EDS Score. This is the start, not the finish.
Try them
Everything’s at focusgts.com/mcp and on GitHub. Pick a server, run one command, and point your AI client at your Adobe stack.