Text2SQL Feedback Collector

When a user tells your agent "that answer looks wrong" or "perfect, thanks" — that feedback now goes somewhere useful. The Feedback Collector captures it directly inside Solid, so you can see how your semantic model is performing in the real world.

What it does

Every Text2SQL response your agent returns can now collect a thumbs up or thumbs down from the user. That signal shows up instantly in the MCP Usage tab, linked to the exact question that triggered it.

Feedback is stored alongside the original Text2SQL response in Solid, so you can always see the answer that triggered each rating.

Over time, this gives you a clear picture of where Solid is doing well and where the model needs tuning — without anyone having to manually review queries or write reports.

How it works in practice

Here's a typical interaction:

User: How many active accounts do we have this month?

Agent: You have 4,821 active accounts this month.

User: That doesn't look right — it's not filtering by the current month.

Agent: Got it, thanks for the feedback. (logs: negative · "Missing month filter")

The agent calls submit_text2sql_feedback in the background. The user doesn't need to do anything differently — they just talk to the agent normally.

Viewing feedback in Solid

Open the MCP Usage tab in your Solid workspace. For each question you'll see:

• Sentiment — positive, negative, or neutral
• Count — how many times this question has been asked and rated
• Comments — any free-text context the user provided

Use this to prioritise which parts of your model to improve first.

Setting it up

Add the submit_text2sql_feedback tool to your agent's MCP connection. Then, when a user gives feedback on an answer, call it with:

That's it. Feedback starts appearing in the MCP Usage tab immediately.