Campaign Monitor for
Agentforce.
Turn email data into useful answers.
Find subscribers, understand campaign activity and preview supported changes before you confirm them.
Campaign Monitor for Agentforce adds packaged topics and actions to Salesforce Agentforce. Your administrator chooses the agent, its instructions and the capabilities available to your team.
Campaign Monitor for Salesforce supplies the product connection and Salesforce context.
Campaign Monitor for Agentforce provides the supported topics and actions.
A customer-owned employee agent, access and a tested set of instructions.
Choose your starting point
- In Salesforce: configure a dedicated employee agent or add the packaged topics to an existing suitable agent. This guide follows that route.
- In Slack: add a Slack channel to the employee agent after it works in Salesforce.
- In an AI app: Hosted MCP is a separate connection to the packaged actions; it does not run the employee agent’s instructions.
The Campaign Monitor Agentforce Starter ↗ includes prerequisite checks, a template and setup scripts. It is configuration you own, not a replacement for the installed packages. If Bitbucket asks for access, contact our team for access to the starter.
Before you start
Use this checklist with your Salesforce administrator before setting up the agent.
Need the base app first? Follow the Campaign Monitor setup guide. For the Agentforce package installation link and currently supported version, contact our team.
Using the starter scripts? Install Salesforce CLI with the Agent plugin, Python 3 and jq. Work in the root of your local starter copy. The read-only preflight checks the org and installed actions; fix every failure before continuing.
Setup walkthrough
Start with one test user and a small, known set of data. Expand a step for instructions and a check before moving on.
01Connect and check Campaign Monitor
Open Campaign Monitor in Salesforce and select the intended client. Confirm a successful sync, then find your test list and subscriber outside the agent.
- Check the Campaign Monitor connection and last successful sync.
- Open a known Contact or Lead and compare their subscriber-list membership.
- Check which clients your test user can see. A client is not the same thing as a subscriber list.
The test user can find the expected record and product data without using the agent.
02Install the add-on and enable Agentforce
The base app is connected to the intended Campaign Monitor client and has completed a successful sync. Install the supported Agentforce extension in the same org.
- In Setup → Installed Packages, confirm Campaign Monitor for Salesforce and Campaign Monitor for Agentforce are present.
- In Setup → Einstein Setup, enable Einstein if required.
- In Setup → Agentforce Agents, enable Agentforce. Einstein alone is not sufficient.
If you need the supported installer, contact our team. Salesforce labels and availability can vary by release and licence.
Both packages are installed in the same org and Agentforce is enabled.
03Give the test user access
Assign an appropriate Campaign Monitor base-app permission set (for example wbsendit__sendIT_Full_Access where the role requires it), plus Campaign Monitor Agent Admin (cm4sfext__Campaign_Monitor_Agent_Admin). Confirm package licensing and visibility of the intended client and Salesforce records.
This starter uses an Employee Agent, which runs as the logged-in user. Do not configure a default Agent User for this route. Access to the new agent is a separate step after it is published.
The user has base-app access, the extension permission and access to the test records.
04Choose and create your agent
Already have a suitable employee agent? Add the packaged Campaign Monitor topics through Add subagent → Browse AgentExchange or the Asset Library. You do not need a second agent or the creation script. Existing-agent instructions ↗
Want a dedicated agent? Use the starter route below. It creates the customer-owned configuration and aligns action inputs and outputs with your installed package.
Dedicated-agent starter: commands and expected result
Get a local copy of the starter repository ↗. Authenticate to the intended test org; replace my-org with your alias throughout.
sf org login web --alias my-org
./scripts/preflight.sh my-orgFix the preflight failures, then generate and validate a local draft:
./scripts/create-campaign-monitor-agent.sh my-org \
--name Example_Campaign_Monitor_Agent \
--label "Example Campaign Monitor Agent" \
--company "Example Organisation" \
--draftKeep --draft on your first run. It creates and validates the local authoring bundle without publishing or activating the employee agent. The generated file is in force-app/main/default/aiAuthoringBundles/<ApiName>/.
You have added topics to your existing agent or created a validated local draft.
05Review instructions and boundaries
- Resolve the exact client, list and person. Use an email address for subscriber operations. Ambiguous matches must prompt a question.
- Separate reads from writes. The first write request produces a no-change preview. Only a later explicit acceptance can apply the same change using its returned confirmation token.
- Keep tokens and internal fields hidden. Preserve the starter’s action contracts and display the packaged safe response.
- Know the limits. Campaign creation/sending, purchases and permanent erasure are not exposed. Sending a configured transactional smart email is a separate supported, confirmed action; it is not a campaign send.
- Respect opt-outs. Membership, stored activity and consent are different. Review downstream effects before subscribing, updating, importing or sending a transactional email.
Customising and preserving your starter
Adapt the company, label, welcome message and tone. Preserve packaged action targets, input/output schemas, employee-agent execution and the product’s confirmation rules.
sf agent validate authoring-bundle --json \
--api-name Example_Campaign_Monitor_Agent -o my-orgDo not regenerate over manual edits. The creation script overwrites the local template. Keep your customised bundle, validate it and follow the repository’s direct publish instructions. Package updates do not automatically rewrite your customer-owned agent.
Customisation guidance ↗The available actions match the team’s purpose and the starter’s safeguards remain intact.
06Preview, cancel and verify
Test a client/list lookup and a named campaign summary. Then preview a refresh of one test subscriber: cancel it, verify nothing changed, and try again with an explicit confirmation. Test an ambiguous name, missing client access and an unsupported campaign-send request.
Testing the draft and inspecting results
Follow the starter’s testing route ↗ for your installed Salesforce tools. If you need a published version in Builder, use the same creation command with --no-activate instead of --draft to publish it without activation.
Inspect the action trace: confirm the intended packaged action ran, its inputs match the request and the returned result supports the answer. Check the underlying record or file separately. Do not treat a timeout as proof that no action occurred.
Read results match the source, cancelling makes no change and a confirmed test action produces a verified result.
07Publish, grant access and retest
Once your administrator has reviewed the tested configuration, publish and activate the intended version.
Starter publishing options
Add --publish to the same creation command to publish and activate. The Campaign Monitor script validates only by default; --no-activate publishes an inactive version.
If you edited the generated bundle, use the repository’s direct validate/publish route instead of regenerating it. Keep the existing API name when intentionally creating a new version.
- Assign the generated <Agent label> Access permission set and confirm its Agent Access section includes the agent. Keep the base-app and Campaign Monitor Agent Admin permissions from step 3.
- Refresh the user’s session after permission changes. Open Agentforce in Salesforce and select the configured agent.
- Repeat a read-only test as an intended end user, not just as an administrator.
- Give colleagues a few precise sample prompts and a route to their Salesforce administrator if results are missing.
The intended users can select the active agent and get the expected results with their own access.
What the agent can do
Your installed package version, configured actions and user access determine which requests are available.
Clients and subscribers
Find accessible clients, subscriber lists and a person’s subscription information.
Campaigns and engagement
Summarise campaigns and journeys, inspect email history and prepare meeting context.
Previewed changes
Refresh or update a subscriber, manage list membership, run a saved import or send an existing transactional smart email with confirmation.
Interactive examples
Work through a read-only question, a useful follow-up and a confirmed change. These examples teach the sequence; they do not connect to your org.
Find the correct subscriber
Before you try it: An accessible client and an existing test subscriber.
Name the person precisely
Check the match and the source
Sam is on Northstar Updates (Demo), with a subscribed status in the synced Salesforce data.
The result should identify its source and any freshness limitation. If the client or person is ambiguous, expect clarification.
Verify the membership
Open the same subscriber’s membership in Salesforce. Compare the client, list and status, then check the last sync. Request a supported live lookup if current Campaign Monitor status is needed. A read must not subscribe anyone.
Cancelled — no change made
The example stops at the preview. In your org, verify the target record or file if you are unsure whether an action ran.
Understand a campaign summary
Before you try it: A named campaign that the test user can access.
Ask about one campaign
Read the campaign context
Northstar Autumn Update (Demo)
120 recipients · 60% opens · 20% clicks.
These illustrative totals describe the campaign, not Sam’s individual activity.
Compare like for like
Open Campaign Details and compare the same campaign, reporting period and data source. Use the person’s email history for person-level questions. Missing synced activity does not prove no engagement.
Cancelled — no change made
The example stops at the preview. In your org, verify the target record or file if you are unsure whether an action ran.
Refresh one subscriber with confirmation
Before you try it: A disposable test subscriber in an accessible test client and list.
Request a specific refresh
Review the no-change preview
Refresh sam.taylor@example.com in Northstar Updates (Demo) for Northstar Demo?
No change has been made. Review the action’s exact preview before accepting.
These buttons only change the illustration. They do not update your connected systems.
Wait for a successful action result
The real agent must use the token from the matching preview and show the returned safe response. Changing the person, list or client requires a fresh preview.
Verify the refreshed data
Reopen the test subscriber and compare the refreshed fields with Campaign Monitor. A refresh retrieves source data; it is not a request to change subscription consent or send an email. Inspect the actual record before retrying a failed or timed-out request.
Cancelled — no change made
The example stops at the preview. In your org, verify the target record or file if you are unsure whether an action ran.
Illustrative responses, not live product output. Exact wording and available results depend on the installed actions and your configuration.
See it in Salesforce
Open an image to inspect the product view at full size. Use the examples above to practise the conversation separately.

Troubleshooting
| What you see | What to check |
|---|---|
| Preflight fails or topics are missing | Check the org alias, CLI Agent plugin, Python 3, jq and installed package versions. Fix the reported failures rather than assuming a fixed action count. |
| The agent is missing from the Salesforce panel | A local draft is not active. Check the published version, activation and the user’s generated Agent Access permission set. Refresh the session. |
| A user cannot find known data | Check their package licence where required, product permissions, extension permission and record access. Check the intended client/organisation and the last successful sync. |
| The answer does not match the source | Confirm the exact record, file or event and the action that ran. Compare the same date range and synced/live source. |
| A requested change is blocked | Review the exact preview and confirmation, permissions and supported operation. A changed target needs a new preview where applicable. Do not bypass the confirmation rules. |
| A request times out or gives no clear outcome | Inspect the source record or file before retrying. Do not assume a timeout means nothing happened. |
| Validation reports a property or schema error | A hand-pasted template can differ from your installed package. Preserve manual edits, then use the starter’s alignment process and validate again. |
Still stuck? contact our team with package versions, the failed step, error text and whether one user or everyone is affected. Leave out passwords, OAuth tokens and other credentials.
Common questions
Do I need to create another agent?
Does --draft publish the employee agent?
Can I use this in Slack?
Can I use an AI app through MCP?
Will a package update overwrite my agent?
Are all answers live from Campaign Monitor?
Can the agent send a campaign?
Does a membership or activity record establish consent?
Invocable actions
Explore the related Campaign Monitor guide.

