Test MCP capabilities

This guide gives you a structured way to exercise the Loomi Connect MCP server against a real workspace. Each section targets a different kind of question such as workspace setup, customers, marketing activity, performance, or shopper experiences. Every section includes single-turn prompts you can run in one shot and agentic prompts that chain multiple tools across an investigation.

Before you start

You need:

• The Loomi Connect MCP server connected to your AI application. See Get started with MCP and Authenticate MCP connection.
• An authenticated session against a workspace with at least one active production project — ideally one with live campaigns, scenarios, and analytics.
• A reader-level Bloomreach IAM role on that project. PII fields appear masked (******) for users without the required permissions/roles. This is expected.

Each section below has two prompt types:

• Single-turn prompts: Call one or two tools and return a direct answer. Use these to try different capabilies on the fly.
• Agentic prompts: Ask your AI application or agent to chain several tools across a multi-step investigation. Use these to showcase the server's real value.

Replace any <angle-bracket placeholder> with a real value from your workspace before sending.

Anchor each session with a prompt that resolves the project once:

Connect to <customer or project name>'s workspace and give me a quick
overview of what's configured there.

Subsequent prompts reuse that context. For a wider view, run a fan-out across the whole organization:

List all projects in my organization with their KPI snapshots, and flag
any that look unusually quiet or unusually busy.

This is a fan-out call and may take longer on organizations with many projects.

Understand a workspace

Use these prompts to get oriented in a new project — its data model, settings, and shape.

Single-turn prompts

What workspaces and projects exist under <organization name>?

Give me the KPI snapshot for project <project name>: customer count,
event count, and active campaign counts by channel.

What identifiers are used to recognize customers in this project?

List the customer properties tracked here. Which ones look custom and
which ones are out-of-the-box?

What events are tracked, and what properties does each event carry?

What consent categories and frequency policies are configured?

List the voucher pools in this project. Which ones are running low
on available codes?

Agentic prompts

I'm new to <customer name>'s account. Walk me through the org, list
every project, classify them as production or test based on KPI volume,
and end with a short description of what each production project
appears to be used for.

Run a governance audit on <project name>. Check consent categories,
frequency policies, project variables, API triggers, and the user
access list. Flag anything that looks risky.

Investigate a customer

These prompts are a good stand-in for the kind of question an internal team member might ask when helping a specific customer — translating a natural-language question into the underlying properties, events, and prediction scores.

Single-turn prompts

Look up the customer with email <customer email> and show me their
key properties.

List the events <customer email> has triggered in the last 30 days.

What segmentations exist in this project, and what segments do they
contain?

How is the <segment name> segment defined? Walk me through the criteria.

What ML predictions are configured — churn, LTV, propensity? Show me
the target event for each.

Agentic prompts

Build a profile of <customer email>. Pull their properties, recent
events, and any prediction scores attached to their account. Write a
short summary of who this customer is, what they've been doing on
the site, and where they sit in any defined segments.

Customer event history is capped at the last 365 days. Older events aren't accessible.

Review the audience strategy in <project name>. List every segmentation
and autosegment, summarize what each represents, and identify overlaps
or gaps. Check which campaigns reference each segment via customer
filters.

Review marketing activity

These prompts cover campaigns, scenarios, experiments, and the catalogs and recommendations that power personalization.

Single-turn prompts

What email, SMS, banner, in-app, and survey campaigns are active
right now?

Show me the configuration of the <campaign name> campaign — its
trigger, target audience, and split variants.

What's on the campaign calendar for the next two weeks?

Show me the flow of the <scenario name> scenario as a node map. What
trigger starts it, what branches exist, and what does each branch send?

What web personalization experiments are currently running?

What catalogs are configured, and what do they appear to be used for?

List the recommendation engines. Which catalog does each one read from?

Agentic prompts

Give me a state-of-marketing report for <project name>. Across every
channel — email, SMS, banners, in-app, surveys, experiments — list
what's active, group by initiative, and flag any campaign that's been
paused or hasn't been edited in 90 days.

Diagnose the <scenario name> scenario. Map every node, identify each
branch and exit point, and tell me where customers could be dropping
out unintentionally.

A 60-second cache on scenario node calls means repeated drill-ins against the same scenario don't trigger extra API requests.

Audit the catalog and recommendation setup. For each catalog, summarize
its purpose and field schema, list the recommendation engines that
depend on it, and check the most recent sync jobs for failures.

Analyze performance

These prompts cover the analytics building blocks — expressions, aggregates, event segmentations — and the saved analyses, dashboards, and ad-hoc queries that run on top of them.

Single-turn prompts

List the aggregates defined in this project. Which ones look like
LTV, recency, or frequency metrics?

Show me the formula behind the <expression name> expression.

List the dashboards in this project. What KPIs is each one tracking?

Show me the saved funnels and summarize what each one measures.

Run an ad-hoc query: sum purchase revenue by product category in the
last 30 days, top 10.

Agentic prompts

Map out the analytics building blocks in <project name>. Group
aggregates and expressions by what they compute — value, engagement,
recency, lifecycle stage — and call out any that look duplicated or
unused.

I want to understand revenue performance over the last 90 days. Pull
KPIs from the project overview, summarize what existing dashboards and
funnels already cover, then run ad-hoc EQL queries to fill the gaps.
Present the results as a single brief.

Ad-hoc EQL queries can take 10 to 30 seconds on large datasets.

Power shopper experiences

These prompts cover the Search side of Loomi Connect — search configuration, merchandising, autosuggest, and the product catalog insights that shape shopper-facing experiences.

Single-turn prompts

Show me the current search configuration: ranking rules, redirects,
and synonym sets.

List the active merchandising rules. Which categories or campaigns do
they apply to?

Show me the category tree and how products are distributed across it.

What autosuggest configurations are active?

Show me product grid insights — which products are over-indexed in
search but under-converting?

What A/B tests are currently running on search or merchandising?

Agentic prompts

Audit the search and merchandising setup in <project name>. Summarize
the active configuration, list any merchandising rules grouped by
campaign or category, and flag products that appear to be misranked
based on conversion data.

Translate this shopper question into a search response: "<sample
shopper query>". Show me which configurations and personalization
signals shape the result.

Use the built-in workflow guides

Loomi Connect ships eight Markdown guides as MCP resources. They load automatically into your AI application's context, so the model already knows how to navigate the hierarchy, look up customers, work with analytics building blocks, explore scenarios, and run ad-hoc EQL queries.

You don't need to enable them, but you can ask your AI application to summarize what it knows:

What built-in Loomi Connect workflow guides do you have access to?
Summarize each one in a sentence.

For a deeper test, point the model at a guide and a project together:

Using the analytics-objects workflow guide as reference, walk me
through how aggregates, expressions, and event segmentations compose
into customer attributes in <project name>. Use real examples from
this project.

This combines a workflow guide and live tool calls in one response — a good showcase of how the resources and tools work together.

Run multi-area workflows

The most demanding test is a single prompt that crosses multiple capability areas. These show what an agent can do that a UI session can't — answering "what happened" and "what should we try next" in the same conversation, instead of jumping between dashboards, exports, and review meetings.

New customer onboarding review

I'm preparing for a kickoff call with <customer name>. Pull together
a brief covering workspace structure, data model maturity, audience
strategy, campaign mix by channel, automation scenarios, analytics
depth, and any governance risks. Cite specific configurations where
it strengthens the point.

Cross-channel campaign audit

Across every channel in <project name>, identify campaigns that
target overlapping audiences. For each overlap, list the campaigns,
the shared customer filter, and any frequency policy that should be
limiting exposure. End with a prioritized list of overlaps to review.

Lifecycle gap analysis

Map every active scenario in <project name> to a customer lifecycle
stage — acquisition, onboarding, activation, retention, win-back,
churn. Tell me which stages are well covered and which have no
scenario at all. Use segmentations and prediction models as signal.

Search-to-marketing loop

Find products that under-perform in on-site search and check whether
any active email or banner campaigns currently promote them. Suggest
where the merchandising configuration or campaign targeting might be
working against each other.

These chains can call dozens of tools in a single response. The model takes longer to complete them, and may ask clarifying questions partway through on large workspaces.

For limitations and known constraints, see Loomi Connect limitations.

Provide feedback

If a prompt produces a wrong, empty, or confusing result, your AI application can record it without leaving the conversation:

That answer wasn't useful. Submit feedback to Loomi Connect with a
short note about what was missing or incorrect.

Feedback shapes which tools and guides land next.