Feedback collection is a feature of MCP distribution. When enabled, consumers using your context in an AI coding tool can submit feedback tied directly to that context. Submitted feedback appears in Supernova where admins can review, resolve, or delete it.
Enable feedback collection
Feedback collection is configured per context.
- Open the context.
- Go to Distribution → MCP.
- Enable Feedback collection.
Once enabled, the collect_agent_feedback tool becomes available to anyone connected to this context’s MCP endpoint.
How consumers submit feedback
There are two ways feedback gets submitted: manually by the consumer, or automatically via a skill installed in the repository.
Manual submission
Consumers can invoke the collect_agent_feedback tool directly in their AI coding tool at any time. Use this to report a specific issue — a missing token, outdated documentation, incorrect component guidance.
When submitting manually, the more specific the feedback the better. Useful feedback includes:
- What the agent was asked to do
- What the context returned or didn't return
- What the correct behavior should be
Vague feedback like “tokens are wrong” is harder to act on than “the spacing tokens in the button component don't match what's in Figma.”
Automatic submission via skill
Automatic collection uses a skill that monitors AI coding sessions and submits feedback when it detects signals of context quality issues — such as retrieval misses, conflicting context, or token drift.
To set up automatic collection:
- In the MCP card under Distribution, expand Automatic collection.
- Run the command shown to install the skill:
npx skills add supernova-studio/agent-skills --skill capture-frontend-feedback
- Add the provided snippet to your persistent instructions file (agents.md, CLAUDE.md, or .cursorrules).
The skill governs when to capture — when a session touches design systems, styling, tokens, components, or accessibility, and a capture signal fires. The MCP call format, fields, and categories are defined inside the skill file itself.
Review feedback in Supernova
Submitted feedback is visible in the Feedback tab of the context.
Each feedback item shows:
- The category of issue reported
- The agent and user exchange that triggered the feedback
- Submission timestamp
To resolve feedback: Open the item and click Resolve. Resolved items remain visible in the list — they are not deleted.
To delete feedback: Open the item and click Delete. Deletion is permanent.
Use the arrows in the feedback drawer to move between items without closing and reopening drawer.
Best practices
- Review feedback on a regular cadence — weekly is a reasonable starting point for active contexts. Feedback accumulates quickly once teams are actively using the MCP connection
- Prioritize feedback in categories like retrieval_miss and conflicting_context — these indicate the context is missing data or containing contradictory information that directly affects agent output quality.
- After making changes to context scope based on feedback, republish the context and monitor whether the same category of feedback recurs.
- Don't wait for feedback volume to grow before acting. A single user_correction item pointing to a specific token group or documentation page is often enough to identify and fix a real gap.