// the find
linshenkx/prompt-optimizer
An AI prompt optimizer for writing better prompts and getting better AI results.
A TypeScript tool for iteratively improving LLM prompts — you paste a prompt, pick a model, and it rewrites the prompt with the goal of getting better outputs. Ships as a web app, desktop app, Chrome extension, and Docker container. The audience is anyone who writes prompts regularly and wants a structured way to test variations rather than tweaking manually.
The client-side-only architecture is genuinely correct for this use case — your API keys never leave the browser, which matters for a tool that handles production credentials. Multi-round iterative optimization (rather than a one-shot rewrite) is the right model, since a single pass rarely catches everything. The MCP server integration means you can call it from Claude Desktop or other MCP clients programmatically, which extends its usefulness beyond the UI. Variable substitution in templates is well-thought-out for teams building parameterized prompts across different contexts.
The CORS problem is real and acknowledged but the fix (use the desktop app) is a workaround, not a solution — any web deployment hitting local Ollama or strict-CORS commercial APIs is broken by default. AGPL-3.0 is a meaningful constraint that a lot of companies will hit: if you embed this in an internal deployment, you're technically obligated to open-source your changes. The optimization quality depends entirely on which model you use to do the optimizing, and there's no guidance on which models actually produce better results versus which ones just produce longer prompts. Version history for saved prompts exists but there's no diff view, so comparing v1 and v5 of a prompt requires manual inspection.