Capabilities and Limitations

Authoritative coverage for Ternary MCP, including FOCUS-only queries, budget snapshots, dashboard comparison limits, and host-side period comparison.

This page is the reference for what Ternary MCP currently exposes and where important limitations apply.

For a less technical introduction, see What you can do with Ternary MCP.

Current product coverage

Cost and usage queries

Ternary MCP can run ad-hoc cost and usage queries with:

• Date range
• Grouping dimensions
• Measures
• Filters
• Time precision

The response includes:

• Returned groups
• An exact server-computed total for the requested range
• Information about how much of that total is represented by the returned rows

Data source limitation

Ad-hoc MCP cost queries currently run against:

FOCUS_BILLING

Other Ternary data sources are not currently exposed through this path, including:

• Kubernetes
• Carbon-footprint data
• Custom datasets

Do not assume that an MCP cost-query result represents every dataset available elsewhere in Ternary.

Saved reports

Ternary MCP can:

• List saved reports
• Search reports by name
• Retrieve report rows
• Create a new private report from a cost analysis

A report created through MCP:

• Is private to the authenticated user
• Appears in saved reports
• Can only be created when the user explicitly asks to save the analysis

MCP cannot currently update or delete the saved report after creation.

Budgets

Ternary MCP can list budgets with figures shown in the Budgets experience, including:

• Amount
• Spend to date
• Projected month-end spend
• Health status
• Period of the latest evaluation

Health status can include:

• On track
• At risk
• Over budget
• Not yet evaluated

Budget limitations

Budget data is a snapshot from the budget's last evaluation.

Important implications:

• The reported period can predate the current month
• Budgets are currently monthly in this MCP contract
• MCP does not provide a month-by-month budget history
• The budget projection is not the same thing as the Ternary Forecasting product

For finance-critical questions, check the budget period and evaluation context returned with the result.

Dashboards

Ternary MCP can:

• List dashboards
• Search dashboards by name
• Retrieve dashboard composition
• Resolve the reports or budgets represented by widgets
• Compare supported dashboard report widgets across two periods

Dashboard comparison

Dashboard comparison can return, per supported widget:

• Current value
• Prior value
• Absolute change
• Percentage change
• Groups with the largest movements

By default, the dashboard comparison uses the last seven full days and the seven full days before them.

Up to 15 report widgets are compared per call.

If a widget cannot be compared honestly, Ternary returns the reason instead of manufacturing a figure.

Examples include:

• Variance reports
• Non-summable measures such as averages or maximums
• Widgets whose data could not be read

Budget widgets pass through their existing evaluated status rather than being recomputed for the comparison window.

Invoice-month limitation

If a report normally accrues by invoice month, dashboard comparison currently evaluates it over a usage-timestamp window.

The result therefore does not use the report's normal invoice-month accrual basis. Affected widgets disclose this condition.

Cost alerts

Ternary MCP can retrieve cost alerts fired by Ternary alert rules.

It includes:

• Anomaly detections
• Threshold breaches
• Observed value
• Evaluation baseline
• Alert status
• Information about the rule

The default alert window is the last 30 days.

Cost-alert limitations

MCP reads detections already made by Ternary.

It does not independently scan arbitrary cost data and decide that a value is anomalous.

The rule information shown with an alert represents the rule's current configuration, not necessarily its exact historical configuration at the moment an older alert fired.

Alert status is a human triage state from Ternary. It does not change automatically because the underlying cost later changes.

Savings recommendations

Ternary MCP can list current savings recommendations, largest estimated monthly savings first.

The MCP surface covers supported optimization recommendations such as:

• Idle resources
• Rightsizing
• Other currently applicable optimization types exposed by the Recommendations experience

Recommendations can be filtered by supported cloud provider and state. The current recommendation view reflects currently applicable recommendations from the last 30 days.

Recommendation limitations

Commitment inventory and commitment recommendations are not currently exposed through MCP. Savings recommendations should not be treated as an explanation for a budget variance unless the underlying evidence supports that conclusion.

AI spend against revenue

Ternary MCP can weigh AI spend against revenue over a time window, for the whole tenant and for each value of one Custom Label, such as each product or team.

The result can include:

• AI cost
• Revenue
• The ratios between AI cost and revenue

The default window is the last 30 days.

AI ROI limitations

AI cost comes from Ternary's own AI classification of each billing row. It is not based on service names or tags.

Revenue comes from the Revenue field of a bring-your-own-data (BYOD) import. Revenue must be in Ternary for the requested window before a ratio can be calculated.

A per-product or per-team ratio also needs a billing rule that assigns the Custom Label to your revenue rows.

Until both are in place, Ternary returns AI cost and names the missing piece instead of estimating a ratio.

Each question can use one Custom Label.

General period comparison

Ternary has a server-computed dashboard comparison capability. General ad-hoc cost comparison is less deterministic today.

For a question such as:

Why did AWS spend increase last month?

an AI assistant may need to:

1. Determine the intended periods
2. Run multiple cost queries
3. Calculate the difference
4. Rank the returned groups
5. Describe the contributors

That arithmetic and interpretation may happen in the AI host rather than in a single Ternary-owned semantic operation.

For finance-critical comparisons:

• Define the period explicitly
• Name the comparison baseline
• Keep the measure and filters consistent
• Treat a contributor as a contributor, not a proven root cause

When an existing dashboard represents the financial view you need, prefer dashboard comparison because Ternary computes the current and prior values server-side.

Permissions

MCP results follow the authenticated user's Ternary permissions. A missing object can mean that the user does not have access to it. See Security and permissions.

Client-specific limitations

AI clients do not all expose MCP in the same way. For example, ChatGPT Deep research currently uses only the MCP search and fetch contract and cannot run the complete Ternary tool set. See the setup guide for your client before assuming that every Ternary capability is available in every mode.