🧠PromptSuggestion · Smart Dislike MCP Server
Turn a thumbs down into a sharper prompt.
This MCP server reads the conversation behind a disliked LLM response, identifies what went wrong, and gives you one focused follow-up prompt you can send immediately.
How it feels in chat
Suggested prompt:
Explain what Docker is and what problem it solves, using a simple analogy. Then give me two concrete examples of how a beginner would use it.
The MCP server always returns one best prompt.
Watch the Demo
A quick preview showing how Smart Dislike turns a bad answer into a perfect follow-up prompt.
Core idea
One solid follow-up prompt
Instead of five vague suggestions, you get one carefully rewritten prompt tuned to your last request.
Context aware
Understands the chat
You pass recent messages + an optional comment. The server reads the dynamic and fixes what went wrong.
Plug & play
Works wherever MCP works
FastMCP cloud, Chrome extension, or local Claude Desktop setup.
For builders
FastMCP / Dev mode
Connect the hosted MCP server URL in your LLM's developer mode, then call the tool whenever you need a better prompt.
Everyday chat
Chrome extension
Adds a Smart Dislike button on top of the normal thumbs-down in the browser, without touching the model UI.
Local workflow
Claude desktop
Run the MCP server locally and wire it into Claude desktop for fully local experimentation.
How PromptSuggestion thinks
Three quick steps from 👎 to a better prompt
1. Capture the moment of frustration
You dislike an answer and pass a small window of recent messages + optional comment.
No need to explain everything again.
2. Analyze intent vs. reply
It infers what you really wanted and why the assistant’s response fell short.
3. Return a sharper prompt
You get one clean, first-person prompt plus a small summary + confidence score.