Community Manager

Workfront Fusion MCP and Adobe Coworker integrations are now live

2 replies
157 views
Translate All

Overview

Today we’ve turned on the Fusion MCP server and integrated Fusion with Adobe Coworker. That means you can now interact with your Fusion organization — scenarios, executions, connections, webhooks, data stores, activity logs, and more — via natural‑language in Coworker or any MCP‑compatible agentic platform that supports remote (Streamable HTTP) MCP servers with OAuth.

What’s new

Fusion MCP server is available at https://mcp.fusion.adobe.com/mcp for custom MCP integrations (Claude, ChatGPT, Copilot, Gemini, Cursor, VS Code, and others).
Adobe Coworker now includes built‑in Fusion support (standalone Coworker and Coworker in the Fusion right rail). No custom MCP URL or OAuth app registration required for Coworker.
All MCP tools described in the Fusion MCP tools reference are exposed to connected agents and follow the same read/write permissions as your Fusion role and team membership.

How to get started

Choose the surface you want to use:

Coworker (recommended) — Open Coworker (standalone) or the Coworker panel in the Fusion right rail and ask Fusion‑related questions. Coworker uses your Adobe ID and Fusion permissions automatically. To test Coworker integrations: Customization > Integrations > find fusion-mcp > Test.
Claude — Add Fusion as a custom connector and point it at https://mcp.fusion.adobe.com/mcp. (Claude Team/Enterprise owners can add custom connectors.)
ChatGPT / other MCP clients — Add a custom MCP server of type Streamable HTTP and use https://mcp.fusion.adobe.com/mcp. Authenticate via OAuth when prompted.
Custom agents — Connect directly to the MCP server URL from your MCP‑compatible agent implementation.

Permissions and safety

The agent performs actions using your Adobe ID and your Fusion role/team permissions — it cannot access anything you don’t already have permission to view or change.
Tools are labeled Read or Write; destructive operations always require explicit confirmation.
Connection and key tools return metadata only (name, type, scopes, expiration). They do not expose secret values or credentials.

Admin checklist

Ensure users and organizations are signed in with Adobe IDs (IMS).
Enable MCP access for organizations that want agent integrations.
Review and set Read vs Write MCP tool permissions for your teams.
Confirm connected app callback URLs if you add custom OAuth apps for non‑Coworker integrations.

Troubleshooting (quick)

No data visible? — Ask the agent to list organizations and switch to the correct organization (tools: fusion_orgs_list, fusion_orgs_set).

Where to find documentation

See Experience League for full Salesforce module reference and connection field details:

Feedback and support

If you have questions, find an issue, or want to request enhancements (for example, tailored tone templates for Coworker, additional safety confirmations, or org‑level presets), please:

Send feedback straight from your AI assistant with the fusion_feedback_submit tool. Ask it to report a bug, flag a wrong result, request a feature or share praise (for example, "Send feedback to the Fusion team: the scenario list is missing folder names").
Post in the Experience League Community thread for Fusion MCP/Coworker.
Open a support ticket via the Admin Console Support portal for migration or access issues.

We’re excited to make Fusion accessible via conversational, agentic workflows. Try Coworker or add the MCP server to your preferred agent and let us know how it goes.

Ross_Barton

Adobe Champion

This is a big one, thanks for getting it out the door. Splitting Read and Write permissions by team is exactly the kind of control admins have been asking for, and returning only metadata from the connection tools (no secrets) is a smart default.

A few questions I expect will come up from a lot of us:

Cost: Do MCP calls use Fusion operations, Adobe AI credits, or neither? Does it work any differently through Coworker than through an outside client like Claude or ChatGPT?
Write scope: Can the agent create or edit scenarios and blueprints, or is write limited to things like running, scheduling, and turning scenarios on and off?
Audit trail: When an agent makes a change, does the activity log show it as an MCP or agent action, or just as the user?
Legacy Fusion: Does this work on legacy Fusion orgs, or only the new experience?
Execution data: Execution logs and data stores can hold sensitive payload data. Can admins limit those tools separately, or is Read all or nothing?

The use case I'm most excited about is scenario triage: asking an agent why something failed last night, or which connections expire in the next 30 days, without digging through execution history. I'm going to connect it to Claude and start with a read-only inventory of our org. I'll share what I learn here.

Translate

DavidGevorgyan

Author

Community Manager

Thanks, Ross, and great questions. Taking them in order:

Cost: The Fusion MCP server is free to use, including from external clients like Claude or ChatGPT. Coworker is metered in AI Credits; details are in the Coworker product description. Today the Fusion capabilities in Coworker and via MCP are the same. We plan to add extended agents in Coworker for more advanced use cases later this year.

Write scope: It goes beyond run, schedule, and on/off. Write tools let the agent create scenarios from a blueprint, clone them (including across teams with connection mapping), update name, description, folder, scheduling, or active state, run once, and delete (deleted scenarios can be restored). It can also manage folders and delete queued webhook events. There's no tool today for editing the modules, but many clients are capable of doing this by themself and later the agent can validate a blueprint without saving it.

Audit trail: The MCP server is a wrapper around the Fusion API and acts with your own credentials, so changes are logged in the audit trail like any other action you take, attributed to your user. You can also query the activity log through the agent, for example "who deleted scenarios in the last 7 days?" Distinguishing agent-initiated actions from manual ones is a good point. We'll evaluate it, thanks for the feedback.

Legacy Fusion: As of October 1, all Fusion organizations have been migrated to IMS, so every org can use the Fusion MCP server. Coworker access depends on your entitlement to Coworker.

Execution data: Access is user-scoped. The MCP server acts with your own Adobe ID and Fusion role and team permissions, so if you can see execution logs, data store records, or webhook payloads in the browser, you can access them through MCP too, and if you can't, the agent can't either. Connection and key tools return metadata only, never secret values.

Scenario triage is one of the use cases we're most excited about, and the connection-expiry question works well there. Please share what you learn. You can also send feedback straight from the agent with the fusion_feedback_submit tool.