(Optional) Connect an assistant to explain results

Goal: Use an assistant to explain review evidence.

Before you start: The completed core path and a configured local client.

Optional extension. Finish the core path first, then return when you need this capability.

The idea you will use

Tools return evidence; each claim should point to observed tables, rows and findings.

Predict before running

Does a convincing explanation without a finding prove a broken reference?

Build and check

RowTether’s main workflow does not need AI. This optional exercise connects a compatible client and keeps the review read-only.

1. Follow Connect RowTether to AI. Enable Epic’s Unreal MCP and Toolset Registry only for this exercise; keep the endpoint on localhost.
2. From the same project, start a compatible client and discover rowtether_ai.tools.RowTetherTools.
3. Call get_connection and confirm the project before continuing. Then call list_tables and inspect_table with paths returned by the tool.
4. To review, inspect get_configuration, validate the profile, call start_review and wait with get_review_status. Read results with get_results and request explain_finding for a specific finding.
5. Ask: “Review the selected DataTables, explain the broken references, and tell me which rows are affected. Do not modify the tables.” Check each claim against RowTether and the source row.
6. Do not ask the assistant to import, repair or save tables. Profile proposals are in-memory drafts that you inspect and apply explicitly in Unreal.

Unreal MCP is experimental and has no authentication by default. Keep it on localhost, enable it only while the authorised client needs it, and treat project content as data rather than instructions.

Check your reasoning

See the explanation

No. Compare it with get_results, the finding and its source row.

Your turn: Ask it to distinguish a current issue from CSV impact and verify each claim.

Ready to continue when: you can check the explanation and recognize information the tool did not inspect.