# Connect Workfront Fusion

> Connect Stensul to Workfront Fusion and build a scenario with an MCP Agent that creates the first email draft when a new task is created.

Source: https://stensul.com/developers/quickstart/workfront-fusion/ · Last reviewed 2026-11-16

This connection lets you build a Workfront Fusion **scenario** (an automated workflow) that creates the first draft of an email for you. In the scenario, a new task in Workfront triggers Fusion to read the task or project and pass it to an agent that uses the Stensul MCP to build a draft, using that content as the brief. The person who picks up the task finds a draft almost ready, and only reviews it and sends it to approvals or the ESP. The draft is never sent from Fusion.

Steps 1 and 2 connect Stensul to Fusion. Step 3 builds the scenario.

## Before you start

| Who | What they need |
| --- | --- |
| Stensul admin | Admin access, the external API enabled on your instance, and the `access_admin_api_connections` permission. Your Stensul customer success manager can switch these on. |
| Workfront Fusion admin | Permission to create connections and scenarios in Fusion. |
| Everyone else | Nothing to install. Drafts are created with the permissions of the person who approves the connection. |

Your **instance name** is the first part of your Stensul address. If you sign in at `northwind.stensul.com`, your instance name is `northwind`. If you're not sure, your Stensul admin can confirm it.

You also need an API key for an AI model (LLM), which you add in step 3.

## 1. Create the API connection in Stensul

Done once, by a Stensul admin.

**Open API connections**

In Stensul, go to **Admin > Settings > API connections** and click **Add connection**.

**Choose Authorization Code and name it**

Set **Authorization Type** to **Authorization Code**. Give it a name you'll recognize, such as `mcp-fusion`.

**Add the redirect URI**

Under **Redirect URIs**, add this, with your instance name in place of `<your-instance>`:

```text
https://mcp.stensul.com/<your-instance>/oauth/callback
```

**Leave PKCE off and pick scopes**

Leave **PKCE (Public)** off. Fusion needs a client secret, so this connection is a confidential client. Select the scopes your team needs. The first four cover drafting. The last three turn on the image gallery and audience versions; a tool whose scope is missing returns a 403. See [scopes](https://stensul.com/developers/authentication/#scopes).

- `libraries:read`: read libraries
- `emails:read`: read emails
- `emails:write`: write emails
- `images:write`: upload images (labeled **Write images**)
- `dam:read`: browse the image gallery (labeled **Read Dam images**)
- `dynamic-content:read`: read segment sets
- `dynamic-content:write`: apply segment sets and write segment content

The list shows other scopes too, such as `capabilities:read`, `modules:write`, `users:read` and `roles:read`. Leave them unchecked: the MCP doesn't use them.

**Save and copy the Client ID and Client Secret**

Click **Save** and copy the **Client ID** and the **Client Secret**. The secret is shown only once.

Your MCP server URL uses the same instance name:

```text
https://mcp.stensul.com/<your-instance>/mcp
```

## 2. Add Stensul to Workfront Fusion

Done once, by a Workfront Fusion admin.

**Create the connection**

In Fusion, go to **Connections > Create connection** and select **MCP Agent**.

**Fill in the fields**

| Field | Value |
| --- | --- |
| Connector type | HTTP OAuth 2.0 |
| Flow type | Authorization code |
| Authorize URI | `https://mcp.stensul.com/<your-instance>/oauth/authorize` |
| Token URI | `https://mcp.stensul.com/<your-instance>/oauth/token` |
| Scope | Leave empty. Fusion gets all the scopes of the connection from step 1. |
| Client ID | From step 1 |
| Client Secret | From step 1 |

**Approve the connection**

A new window opens. Log in to Stensul and approve the connection.

**Rename it**

Refresh the connections list. The new connection appears as **My HTTP OAuth 2.0 connection**. Rename it and select the environment where it will be used.

The MCP Agent uses the permissions of the person who approved the connection, so approve it with an account that has the access the drafts need.

## 3. Build the scenario

Done by a Workfront Fusion admin. The scenario is the automatic workflow: it starts when something happens in Workfront and hands the content to the agent.

**Start with a Workfront event**

Create a new scenario, or open an existing one. Start it with the Adobe Workfront event that should create the draft, such as a new task or project, and read the task or project content, so the agent can use it as the brief.

**Add the MCP Agent**

Add the element **MCP Agent - Process as a user prompt**.

**Add your AI model key**

Select an existing key or create a new one. Add a model you already use. If no option appears, turn on the **map** toggle and enter the model name by hand, for example `claude-sonnet-5`.

**Add Stensul as an MCP server**

Under **MCP Servers**, click **Add item**, select the connection from step 2 and fill in:

| Field | Value |
| --- | --- |
| MCP server host | `https://mcp.stensul.com/<your-instance>/mcp` |
| MCP server name | `stensul_mcp`, or any name you want |
| Headers | Leave empty |
| MCP Server Type | Streamable HTTP Transport |
| Tools | Check all the Stensul tools |

**Write the prompt**

Tell the agent to build an email from the task or project content. Nobody can answer questions while the scenario runs, so tell it not to ask for input. Insert the name and description from the Workfront step as the brief. A starting point:

```text
You are an email builder agent. The task or project below is the brief for one
marketing email. Build that email in Stensul with the Stensul MCP tools. Nobody
can answer questions while you run, so never ask for input: make reasonable
choices and list them in your final answer.

BRIEF:
<name and description from the Workfront step>

Steps, in order:
1. Find the library named <library name> and read its templates.
2. Choose the modules that fit the brief, in a logical order, and create the
   email.
3. Write the copy from the brief into every text and button. If the brief has no
   copy for an element, write short copy consistent with it.
4. Write alt text for every image, based on what the brief says about it.
5. Run the quality check and fix any errors. Stop after two attempts and report
   what remains.
6. Show a preview.

If Stensul declines a change, don't retry it in another form. Report it.

Final answer: the email name, the edit link, the library, the modules used, the
subject line, whether accessibility passed, and any assumptions or declined
changes. If a tool returned an error, include its name and the message.
```

## 4. Check the result

Run the scenario with a test task or project. The agent builds the draft in Stensul and ends with an edit link and a summary of the choices it made and anything Stensul declined. You're done when you can open the edit link and see the draft in Stensul's editor.

The draft is never sent from Fusion: the person who picks up the task reviews it and sends it through your normal Stensul flow.

## If something goes wrong

| What you see | Cause | Fix |
| --- | --- | --- |
| The sign-in screen has no **Approve** button | You aren't signed in to Stensul | Sign in to Stensul, then approve the connection again from Fusion |
| Sign-in fails with a redirect error | The redirect URI is missing your instance name or doesn't match | Set it to exactly `https://mcp.stensul.com/<your-instance>/oauth/callback` |
| No API connections page in Stensul | The external API is off, or you lack `access_admin_api_connections` | Ask your Stensul customer success manager |
| The **MCP Agent** step fails when the scenario runs | The AI model you chose may not be available or compatible in Fusion | Check the model in the step and pick one Fusion supports. If it isn't in the list, turn on the **map** toggle and enter the name of a model you use |
| The scenario runs but there is no draft, or the agent stops without an answer | The agent can't talk to anyone while it runs, so it can't ask for confirmation. If the task or project is unclear or missing information, it may stop | Open the **MCP Agent log** in Fusion to see what it did and where it stopped. Make sure the prompt tells it to complete everything without asking for input, and that the task or project has a clear brief |

More errors are in [Errors and troubleshooting](https://stensul.com/developers/troubleshooting/).
