New: Adobe Workfront MCP and Adobe AEM Assets MCP connectors

Two new MCP connectors bring conversational, model-driven automation to Fusion. Instead of wiring together many individual modules, a single MCP module accepts an open-ended instruction and lets an AI model determine which Workfront or AEM operations are needed to fulfill it.

Overview

This release adds two first‑party connectors that speak to Adobe's Model Context Protocol (MCP) server: Adobe Workfront MCP and Adobe AEM MCP. Each connector exposes a single, flexible module you place on a scenario canvas and drive with natural‑language instructions (prompts). The MCP module synthesizes the required API calls, executes them, and returns a consolidated, human‑readable result — reducing the need to build long chains of Get/Filter/Transform/Write modules for many common tasks.

What’s new

• Single conversational module — One module accepts an instruction and lets the MCP model plan and run the necessary Workfront or AEM operations, returning a synthesized answer or structured output.
• Task examples — Ask the module to summarize active projects behind schedule, bulk-annotate assets based on content, find and tag orphaned files, or generate a prioritized list of at‑risk work items — without manually composing multiple connector calls.
• Action restrictions & safe mode — Define an allowlist or denylist of permitted Workfront/AEM operations so unattended scenarios cannot perform unexpected destructive actions (for example, disallow Delete or Update actions and allow only Read/List).
• Default Adobe Managed AI — Out of the box the MCP module uses Adobe Managed AI backed by the claude‑sonnet‑5 model for planning and synthesis.
• Bring your own model (BYOM) — Optionally configure the module to call a different LLM by supplying a model key and credentials. The MCP runtime will route instructions to the model you provide while still enforcing the action restrictions you’ve set.
• Structured and human responses — Receive either a synthesized narrative summary or structured JSON with the model’s chosen actions and results for downstream scenario processing.
• Auditable plan preview — Before executing, optionally request the model to return a proposed step plan (which modules/actions it intends to run) so you can review and approve in interactive or CI/CD flows.

Why this matters

The MCP connectors reduce scenario complexity and development time by moving orchestration logic into a model that can reason across resources and operations. Common benefits include:

• Faster automations — Build fewer modules: one instruction can replace multi‑step flows that previously required Get → Filter → Transform → Write sequences.
• Better human‑readable outputs — The model synthesizes and summarizes results, making outputs easier to consume for stakeholders and downstream notifications.
• Flexible interactions — Non‑developers can drive automation through prompts without deep knowledge of each connector’s API surface.
• Safer unattended runs — Action restrictions and an optional plan preview let Ops and security teams approve or limit what the model may change.
• Model choice — Use Adobe Managed AI for a turnkey experience or plug in a different LLM when you need specific models, fine‑tuning, or data residency requirements.

How to get started

1. Open Fusion and create or edit a scenario.
2. From the connector list, add either the Adobe Workfront MCP or Adobe AEM MCP module to your canvas.
3. Create a connection to the MCP server (Adobe Managed AI is configured by default). If you want to use a different model, provide the model endpoint/key and any required credentials in the connection settings.
4. Author your instruction in the module’s prompt field. Examples:
   • “Find all active projects that are behind schedule and summarize their status, including top three blockers and suggested next steps.”
   • “Identify AEM assets with missing metadata that reference ‘summer campaign’ and add the 'campaign:summer' tag.”
5. Configure allowed actions — select which Workfront/AEM API actions the model may invoke (Read/List/Update/Create/Delete). Enable Safe Mode (Read-only) for discovery scenarios.
6. Map the module’s outputs into downstream modules (for example, send the synthesized summary to Slack, create a Workfront report, or write structured results to a datastore).
7. Test in a non‑production org or team, then publish and schedule or trigger as required.

Behavior and tips

• Start with conservative allowlists — allow Read/List first, validate the model’s step plans, then expand permissions to Create/Update if needed.
• Use clear, explicit prompts. If you require non‑destructive behavior, include it in the instruction (for example, “Do not delete anything; only suggest items to delete.”) and enforce with the connector’s action restrictions.
• When using BYOM, verify the model’s latency and cost characteristics and test prompt reliability in a staging environment before production roll‑out.
• Enable Plan Preview for high‑risk instructions so you can review the model’s intended actions and catch mistakes before they run.
• Combine MCP outputs with existing Fusion modules — for example, use the model’s structured result to feed a bulk Update or to create tickets in your ITSM system.
• Audit logs and activity exports are recommended for governance: pair MCP scenarios with Insights or activity exports so you can trace which automated actions were executed and by which scenario.

Before you start

• Review organizational policies on AI usage, data sharing, and model selection. BYOM may require additional approvals for external model keys.
• If you plan to allow Update or Delete actions, coordinate with Team leads to ensure scenarios run in appropriate test environments first.

Where to find documentation

See Experience League for full connector documentation, prompt examples, and security guidance:

• Experience League — Workfront MCP connector documentation
• Experience League — Workfront MCP documentation
• Experience League — AEM MCP documentation
   

Need help or want to give feedback?

If you have questions, encounter issues, or want to suggest improvements (for example, richer plan visualization, templated prompts for common workflows, bulk operation safety controls, or additional model integrations), please:

• Post a comment on the Experience League article
• Start a thread in the Experience League Community
• Contact Adobe Support through the Admin Console Support portal

Your feedback helps us prioritize future enhancements.