Adobe Fusion MCP modules

Overview

The new Adobe Fusion MCP connector lets you send a plain‑English prompt from a Fusion scenario to the hosted Adobe Fusion Model Context Protocol (MCP) server. An LLM interprets your request, invokes the Fusion tools required to fulfill it, and returns the result plus execution metadata.

What’s new

• Single‑prompt MCP integration — send one natural‑language instruction to the hosted Fusion MCP server directly from a scenario module.
• Optional BYO LLM — use Adobe Managed AI by default, or supply an LLM key (OpenAI, Anthropic, Amazon Bedrock) and model if you prefer a different provider.
• Tool restrictions — restrict which read‑only and write/delete Fusion tools the MCP model may call to enforce least privilege for unattended runs.
• Audit and summary outputs — the module returns a Response, an Audit Trail of tool calls and results, and a Summary (counts and final status) for transparency and downstream logic.

How it works (quick)

1. Add the Adobe Fusion MCP module to your scenario.
2. Create a connection inside the module (click Add → choose Environment and Type → complete Adobe sign‑in/consent).
3. Enter your plain‑English prompt in the module. Optionally choose or add an LLM key if not using Adobe Managed AI.
4. Optionally select Read‑only and/or Write/delete tools to limit what the MCP model may call.
5. Run the module. The hosted MCP model executes the allowed tool calls and returns Response, Audit Trail, and Summary outputs you can map to later modules.

Why this matters

• Faster, conversational automation — non‑technical users can describe outcomes in plain language and let the model orchestrate Fusion tools.
• Safe unattended runs — tool restriction fields let you prevent unwanted write/delete operations in automated scenarios.
• Transparency — Audit Trail and Summary make it easy to review what the model did and to troubleshoot or audit runs.

Getting started

1. Edit a scenario and add the Adobe Fusion MCP module.
2. Create and authorize a connection via the module’s Connection → Add flow.
3. Compose a clear prompt; start with read‑only requests to validate behavior.
4. If desired, add an LLM key to route prompts to OpenAI, Anthropic, or Bedrock instead of Adobe Managed AI.
5. Restrict tools where required, run the module, and inspect Response, Audit Trail, and Summary.

Troubleshooting (quick)

• Connection issues: Ensure the Adobe sign‑in/consent completed and redirects are allowed by your browser/network.
• Unexpected answers: Clarify the prompt, include constraints or examples, and review the Audit Trail to see which tools produced the output.
• Tool blocked: Confirm the required tool is selected in Read‑only or Write/delete lists; otherwise the MCP server will not call it.
• Custom LLM errors: Verify provider key, model name, and that the provider endpoint is reachable and has quota.

Output fields

• Response — the model’s returned text.
• Audit Trail — record of tool calls, arguments, results, and timings.
• Summary — totals for the session and final status for quick checks or conditional logic.

Documentation & support

See the Experience League for full module details and configuration steps: Adobe Fusion MCP module (Experience League). For help or feedback, post in the Experience League Community or open a support ticket via the Admin Console Support portal.

Try the Adobe Fusion MCP connector to add natural‑language orchestration to your Fusion scenarios — and let us know what prompt controls or safety features you’d like next.