Three production-grade Model Context Protocol servers that give Claude, Cursor, ChatGPT Desktop, and GitHub Copilot real access to Adobe Experience Cloud. AEP MCP is the first full-featured server for Adobe Experience Platform. Firefly Services MCP brings generative image and video, the Photoshop API, and the Lightroom API straight into your AI client. EDS MCP covers Adobe Edge Delivery Services (preview, publish, content, metrics). And this is just the start — more Adobe Experience Cloud MCP servers are in active development.
Why this exists
Most Adobe staffing firms hand you a resume and walk away. We build the tools our placements use, and we open-source them so the rest of the Adobe ecosystem benefits.
Model Context Protocol is the standard Anthropic introduced for connecting AI clients to external tools and data. When Claude, Cursor, ChatGPT Desktop, or GitHub Copilot need to actually do something — not just talk about it — an MCP server is what makes that possible. The AI gets a list of typed tools, calls the ones it needs, and gets structured results back.
Adobe ships first-party MCPs for a small number of its products, mostly read-only beta releases with narrow tool surfaces. The rest of Adobe Experience Cloud has no MCP coverage at all. That gap is what Focus GTS is filling. We started with the three layers our customers ask about most — Adobe Experience Platform, Adobe Firefly Services, and Edge Delivery Services — and we're actively building more as Adobe ships new surfaces and our customers hit new gaps.
All three servers are on npm. All three install with a single command. All three run locally as stdio processes — nothing in our infrastructure, no telemetry phoning home, no rate limits. EDS MCP and Firefly Services MCP and AEP MCP are all Apache 2.0 open source. The repos live on GitHub alongside the rest of our open-source Adobe tooling, including EDS Score, the free Edge Delivery Services audit we run for the community.
Package 1 · Adobe Edge Delivery Services
Give your AI agent real control over an Edge Delivery Services site — preview pages, publish content, query live performance metrics, audit the sitemap, and purge CDN cache. Twenty tools. Apache 2.0 licensed. Zero dependencies beyond the MCP SDK.
@focusgts/eds-mcp-server · v0.2.0 · npm
What it does. Connects an AI client to the Adobe Edge Delivery Services admin API and content bus. Your agent can read pages via the .plain.html endpoint, list the page index, trigger preview and publish for one page or a hundred, check live status, purge cache, and query Core Web Vitals from OpTel — all without leaving the AI conversation.
Install for Claude Code
claude mcp add eds \ -e EDS_OWNER=your-org \ -e EDS_REPO=your-site \ -- npx @focusgts/eds-mcp-server
| Tool | Category | What it does |
|---|---|---|
| eds_preview_page | Publishing | Trigger preview for a page so content source changes appear on *.aem.page |
| eds_publish_page | Publishing | Publish a page from preview to the live *.aem.live domain |
| eds_unpublish_page | Publishing | Remove a page from the live site |
| eds_preview_and_publish | Publishing | Preview and publish in one atomic operation |
| eds_get_status | Publishing | Get preview, live, and code-bus status for a resource |
| eds_purge_cache | Publishing | Purge CDN cache for a page path |
| eds_bulk_preview | Bulk ops | Preview multiple pages in one call (up to 100) |
| eds_bulk_publish | Bulk ops | Publish multiple pages in one call (up to 100) |
| eds_get_page | Content | Fetch rendered page content via .plain.html |
| eds_list_pages | Content | Query the site's page index with pagination |
| eds_get_sitemap | Content | Read and parse the live sitemap.xml |
| eds_get_metadata | Content | Fetch metadata for a page (title, description, OG tags) |
| eds_search_content | Content | Search the page index by title, slug, or path pattern |
| eds_get_cwv | Analytics | Query Core Web Vitals (CrUX field data) for any page |
| eds_get_404s | Analytics | Pull recent 404 reports from OpTel |
| eds_list_experiments | Analytics | List active A/B experiments on the site |
| eds_get_config | Config | Read the helix-config.json head and config |
| eds_validate_config | Config | Validate config against EDS schema before commit |
| eds_list_redirects | Config | Read the redirects.json / redirects spreadsheet |
| eds_health_check | Diagnostic | Full site health pass: status, perf, broken links, config drift |
Package 2 · Adobe Experience Platform
The first full-featured Model Context Protocol server for Adobe Experience Platform. Twenty-three tools across schemas, datasets, identities, profiles, segments, sources, destinations, and Query Service — full read and write, working pagination, and structured error codes.
@focusgts/aep-mcp-server · v0.3.0 · npm
Why this matters. Adobe ships an official MCP for Adobe Journey Optimizer, but it's a read-only beta with three tools, broken pagination, and works only against Claude's hosted remote-MCP transport. Nobody — not Adobe, not the community — has shipped a full-featured MCP for Adobe Experience Platform itself. That's the gap this fills. Own the AEP MCP layer and you own schemas, datasets, profiles, identities, segments, sources, destinations, and SQL queries across the entire Adobe Experience Cloud foundation.
Install for Claude Code
claude mcp add aep \ -e AEP_CLIENT_ID=your-client-id \ -e AEP_CLIENT_SECRET=your-client-secret \ -e AEP_ORG_ID=your-ims-org \ -e AEP_SANDBOX_NAME=prod \ -- npx @focusgts/aep-mcp-server
| Tool | Category | What it does |
|---|---|---|
| aep_list_schemas | Schemas | List XDM schemas in the Schema Registry |
| aep_get_schema | Schemas | Fetch a single XDM schema by ID |
| aep_create_schema | Schemas | Create a new XDM schema (write) |
| aep_list_datasets | Datasets | List datasets in the catalog |
| aep_get_dataset | Datasets | Fetch a single dataset by ID |
| aep_create_dataset | Datasets | Create a new dataset bound to a schema (write) |
| aep_list_identity_namespaces | Identities | List identity namespaces |
| aep_get_identity_graph | Identities | Fetch the identity graph for a given identity |
| aep_get_profile | Profiles | Fetch a Real-Time CDP profile by entity ID |
| aep_preview_profile | Profiles | Preview a profile fragment before ingestion |
| aep_list_profile_views | Profiles | List configured profile views in the sandbox |
| aep_delete_profile | Profiles | Delete a profile for GDPR / CCPA compliance (write) |
| aep_list_segments | Segments | List segment definitions in the sandbox |
| aep_get_segment | Segments | Fetch a single segment definition by ID |
| aep_create_segment | Segments | Create a segment definition from PQL (write) |
| aep_evaluate_segment | Segments | Trigger on-demand segment evaluation |
| aep_list_sources | Sources | List configured source connections |
| aep_create_source | Sources | Create a new source connection (write) |
| aep_list_destinations | Destinations | List configured destination connections |
| aep_activate_segment | Destinations | Activate a segment to a destination (write) |
| aep_run_query | Query Service | Execute a SQL query against the data lake |
| aep_list_queries | Query Service | List recent query runs and their status |
| aep_schedule_query | Query Service | Schedule a recurring SQL query (write) |
Package 3 · Adobe Firefly Services
Generate images and video with Adobe Firefly, run Photoshop API operations on PSD templates, and batch-process through the Lightroom API — all from inside your AI client. Eighteen tools across three Adobe creative APIs, on a commercially-safe generative foundation.
@focusgts/firefly-services-mcp · v0.2.3 · npm
One server for the whole Firefly Services suite. The other Adobe Firefly MCP servers we found cover image generation only. This is the first to bring Firefly, the Photoshop API, and the Lightroom API together in a single MCP server — a thin, production-grade adapter over Adobe's official SDKs. Your AI agent can generate a text-to-image with Firefly V3, expand or fill an existing image, composite a product into a generated scene, generate video, replace a smart object in a PSD, remove a background, or apply a Lightroom preset — each from a single tool call, without leaving the editor. The eight Firefly tools are live-validated end-to-end against the real Adobe API; the Photoshop and Lightroom tools are SDK- and mock-validated and write to your own storage bucket.
Install for Claude Code
claude mcp add firefly-services \ -e FIREFLY_SERVICES_CLIENT_ID=your-client-id \ -e FIREFLY_SERVICES_CLIENT_SECRET=your-client-secret \ -- npx @focusgts/firefly-services-mcp
| Tool | API | What it does |
|---|---|---|
| firefly_generate_image | Firefly | Generate an image from a text prompt with Firefly V3 |
| firefly_generate_similar | Firefly | Generate variations of an existing image |
| firefly_expand_image | Firefly | Generatively expand the canvas of an image |
| firefly_fill_image | Firefly | Generative fill / inpaint a masked region |
| firefly_generate_object_composite | Firefly | Composite a product or object into a generated scene |
| firefly_generate_video | Firefly | Generate a video clip from a prompt or image |
| firefly_upload_image | Firefly | Upload an image and get a reference for later calls |
| firefly_check_auth | Firefly | Verify Firefly Services credentials and token |
| photoshop_smart_object_replace | Photoshop | Replace a smart object's contents in a PSD |
| photoshop_document_manifest | Photoshop | Read a PSD's layer structure and manifest |
| photoshop_apply_actions | Photoshop | Run a Photoshop .atn action set on a document |
| photoshop_edit_text | Photoshop | Edit text layer content in a PSD |
| photoshop_apply_edits | Photoshop | Apply a batch of document edits |
| photoshop_remove_background | Photoshop | Remove the background from an image |
| lightroom_apply_preset | Lightroom | Apply a Lightroom preset to an image |
| lightroom_auto_tone | Lightroom | Auto-tone an image (exposure, contrast, color) |
| lightroom_auto_straighten | Lightroom | Auto-straighten an image |
| lightroom_apply_edits | Lightroom | Apply a batch of Lightroom edit parameters |
Honest comparison
Adobe has shipped a small handful of read-only beta MCP servers. None of them target Adobe Experience Platform itself. Here's the apples-to-apples breakdown.
| Capability | Adobe AJO MCP (beta) | @focusgts/aep-mcp-server |
|---|---|---|
| Target product | Journey Optimizer only | Experience Platform (foundation layer) |
| Operations | Read-only | Full CRUD (read + write) |
| Tool count | 3 | 34 across 10 categories |
| Pagination | Broken (first 50 only) | Working (offset / limit / hasMore) |
| Client compatibility | Claude only | Claude, Cursor, ChatGPT, Copilot, any MCP client |
| Transport | Hosted remote | stdio (local, no data leaves your machine) |
| Sandbox support | Yes | Yes (auto-scoped) |
| Error responses | Sometimes 502 silent | Structured AEP_{status} codes |
EDS and AEP are the first two we shipped publicly. We build private MCP servers on engagement for Adobe Workfront, Adobe Analytics, Adobe Target, Adobe Marketo Engage, Adobe Campaign, and custom Adobe I/O integrations. Your team gets the same production-grade scaffolding — OAuth S2S auth, structured errors, working pagination — tuned to your stack.
Scope a custom MCP buildAdobe Silver Solution Partner · 500+ Adobe specialists placed since 2018
Who this is for
Both servers are designed for teams that already run Adobe in production and want their AI client to do more than write boilerplate.
aem.live dashboards.Focus GTS is an Adobe Silver Solution Partner and has placed 500+ Adobe specialists at Fortune 100 companies since 2018. Our staffing relationships are why we know which Adobe workflows are worth wrapping in an MCP and which ones are not. When you hire us for Adobe + AI staffing or Navigator — our managed Adobe service — you get engineers who know these MCP servers cold because we built them.
FAQ
MCP is an open standard for connecting AI clients — Claude, Cursor, ChatGPT Desktop, GitHub Copilot — to external tools and data sources. An MCP server exposes a typed set of tools the AI can call. Focus GTS publishes three MCP servers for Adobe: EDS MCP for Adobe Edge Delivery Services, AEP MCP for Adobe Experience Platform, and Firefly Services MCP for Adobe Firefly, Photoshop, and Lightroom.
EDS MCP Server is fully open source under the Apache 2.0 license — install it, audit it, fork it, ship it in your own products. AEP MCP Server is also open source under the Apache 2.0 license. Both packages are on the public npm registry and require no registration with Focus GTS.
Adobe ships a read-only beta MCP for Adobe Journey Optimizer with three tools, broken pagination, and Claude-only compatibility. Focus GTS AEP MCP is the first full-featured MCP for Adobe Experience Platform itself — 34 tools across 10 categories, full read and write operations, working pagination, OAuth Server-to-Server auth with token caching, structured AEP_{status} error codes, and compatibility with any MCP-compliant client. AEP is the foundation layer that AJO, Customer Journey Analytics, and Real-Time CDP all sit on top of, so an MCP at this layer covers far more ground than a product-specific one.
No. Both servers run as local stdio processes on your machine. The AI client talks to them over standard input / output. They make calls to the Adobe APIs you configure them against — same as any developer using Adobe's REST endpoints — and return results to your AI client. Focus GTS has no servers in the data path and collects no telemetry from your installs.
EDS MCP works in read-only mode with no credentials at all (content, sitemap, metadata, query index). Write tools (preview, publish, cache purge) need an EDS admin API key. Analytics tools need an OpTel domain key. AEP MCP requires an Adobe I/O OAuth Server-to-Server credential with appropriate AEP product profiles attached — client ID, client secret, IMS org ID, and a target sandbox name. The READMEs in each GitHub repo walk through credential setup.
Yes. We build custom MCP servers on engagement for Adobe Workfront, Adobe Analytics, Adobe Target, Adobe Marketo Engage, Adobe Campaign, and custom Adobe I/O integrations. You get the same production-grade scaffolding behind our public AEP, Firefly Services, and EDS servers — OAuth, structured errors, working pagination, sandbox handling — tuned to your stack. New servers are in active development, so the public suite keeps growing too. Contact us to scope a build.
Yes. Our engineers on Navigator and Adobe + AI staffing engagements use these MCP servers in their daily Adobe work, so they ramp immediately. If you want a senior Adobe + AI engineer who can wire MCP into your build pipelines from day one, our Adobe + AI staffing practice is the fastest path.
Open an issue on the relevant GitHub repository: github.com/Focus-GTS/eds-mcp-server, github.com/Focus-GTS/aep-mcp-server, or github.com/Focus-GTS/firefly-services-mcp. We triage issues weekly. PRs welcome on the Apache 2.0 servers (EDS MCP and Firefly Services MCP).
Related Focus GTS work
MCP servers are one slice of what Focus GTS ships into the Adobe ecosystem. The rest of the portfolio is here.
Install the MCP servers in two minutes. Or hire the team that built them.