An MCP feedback tool could file user feature requests — but may mostly collect compliments

hermjohnson · reddit · 2026-07-30

Using an AI assistant to file feature requests sounds neat — but is it useful or just noisy?

The author explores an idea for MCP-connected apps: if a user asks Claude for something the app cannot do, the assistant could offer to send a feature request directly to the product team instead of forcing the user to open a support form.

They note that this already exists in some form at companies like Ramp and in AWS Marketplace’s Claude connector. But after reading AWS’s documentation, the author becomes skeptical about the signal quality:

the tool is encouraged after recommendations, comparisons, or evaluations,
the examples shown are mostly positive feedback,
and the model may end up prompting feedback almost every turn.

The concern is that the system may collect compliments rather than actionable product gaps.

What the author suspects

The post argues that prompt wording and tool descriptions act more like a volume knob than a precise dial: if you try to discourage a model from using a specific feedback tool, you may just reduce tool use across the board rather than tune that one path.

The author has not shipped the feature yet and is looking for real-world reports on whether such a feedback loop actually works, and whether the resulting data is useful or just noisy.

Original post →

An MCP feedback tool could file user feature requests — but may mostly collect compliments

More from coding & agent