For two years, AI assistants could tell your team how to publish an Adobe Experience Manager page. Now they can actually publish it. That shift, from advice to action, is what Model Context Protocol (MCP) makes possible, and it is quietly changing how Adobe teams work day to day.
If you have wired Claude, Cursor, or GitHub Copilot into your workflow and felt the ceiling, where the model understands your stack but can’t touch it, an MCP server is the thing that removes the ceiling. This post explains what an Adobe MCP server actually does, what we built, and why it matters for anyone running AEM or Adobe Experience Platform in production.
What an MCP server actually is
Model Context Protocol is an open standard, introduced by Anthropic in late 2024, for connecting AI clients to external tools and data. The AI client (Claude Desktop, Claude Code, Cursor, ChatGPT Desktop, GitHub Copilot) speaks MCP. An MCP server exposes a set of typed tools: each one a named action with a defined input and output. The AI reads the list of available tools, decides which to call, sends structured arguments, and gets structured results back.
The mental model: an MCP server is a universal adapter between a language model and a real system. Before MCP, every AI-to-tool integration was bespoke. After MCP, any compliant client can drive any compliant server. That is why the ecosystem went from a handful of servers to thousands inside a year.
The important part for Adobe teams: the AI stops being a brilliant intern who can only talk, and becomes one who can also do the work. It can preview a page, query a schema, or run a report, all inside the same conversation.
What an Adobe MCP server lets your AI do
Concretely, here is the difference. Without an Adobe MCP server, you ask Claude “how do I publish this Edge Delivery page?” and it explains the admin API. With one, you say “publish the pricing page and confirm it’s live,” and it calls the preview tool, calls the publish tool, checks status, and reports back, with no context-switch to a dashboard and no copy-pasting curl commands.
The same pattern applies across the Adobe stack:
- Edge Delivery Services: preview and publish pages (one or a hundred), read rendered content, audit the sitemap, query Core Web Vitals, purge CDN cache.
- Adobe Experience Platform: list and create XDM schemas, inspect datasets, fetch a Real-Time CDP profile, build a segment from PQL, run a SQL query against the data lake.
None of this is magic. Every action maps to an Adobe API that already exists. What MCP changes is who calls it: your AI agent, in plain language, in the flow of work, instead of an engineer hand-assembling API requests.
What we built, and the gap we filled
Adobe has started shipping first-party MCP servers. The furthest along is a read-only beta for Adobe Journey Optimizer, a strong signal that Adobe sees where this is going. We built on that momentum and covered the parts of the stack that didn’t have an MCP yet, and we put both on npm:
- EDS MCP Server (
@focusgts/eds-mcp-server): 20 tools for Adobe Edge Delivery Services, covering preview, publish, bulk operations, content reading, performance metrics, and config. Apache 2.0 licensed, zero dependencies beyond the MCP SDK. - AEP MCP Server (
@focusgts/aep-mcp-server): 23 tools across schemas, datasets, identities, profiles, segments, sources, destinations, and Query Service. Full read and write, working pagination, OAuth Server-to-Server auth, structured error codes.
Both install with a single command and run locally as stdio processes. Nothing routes through our infrastructure, no telemetry phones home, and they work with any MCP-compliant client, not just one vendor’s.
See every tool, with copy-paste install commands
Full tool tables, install instructions for Claude Code, Cursor, and VS Code, and an honest side-by-side with Adobe’s own MCP releases are all on one page.
Explore the Adobe MCP servers →Why your AEM team should care
The value isn’t novelty. It’s the collapse of small, repetitive loops that currently eat senior engineers’ time. “Preview this branch, check it for broken links, publish if it’s clean” is a five-minute human task that an agent with an MCP server does in seconds, and it does it the same way every time. Multiply that across a content team shipping dozens of pages a week and the time recovered is real.
It also lowers the floor on who can safely operate the platform. A marketer can ask an AI agent to publish a page and get a correct, auditable result, because the MCP server encodes the right sequence (preview, verify, publish) as tools rather than tribal knowledge. The senior AEM engineer stops being a bottleneck for routine operations and gets their attention back for architecture.
If you run AEM as a content management system at any real scale, this is the difference between an AI that describes your platform and one that helps run it.
The skill this actually requires
Here is the part most teams underestimate: wiring MCP into an Adobe stack well is its own discipline. It sits at the intersection of Adobe platform knowledge (which API does what, how auth and sandboxes work, where the rate limits are) and modern AI engineering: how to structure tools, how agents reason about them, how to keep a human in the loop on write operations. Very few people have both halves yet.
That gap is exactly what our Adobe + AI staffing practice exists to fill: senior engineers who know Adobe Experience Cloud and the AI tooling layer on top of it. The same people build and use these MCP servers in their daily work, so they ramp on your stack immediately.
If you’d rather embed that capability than hire it outright, Navigator, our subscription managed service, puts a senior Adobe + AI engineer into your team within a week, with no SOWs and no procurement cycle.
Start in two minutes
The fastest way to understand what this changes is to install one and try it. The EDS MCP server works in read-only mode with no credentials at all: point it at any public Edge Delivery site and ask your AI client to audit it. From there, the write tools and the full AEP server are a credential away.
The AI assistants your team already uses are about to get a lot more useful inside Adobe. The teams that wire them in first will feel it first. Start with the Adobe MCP servers →