The useful part is not that it can rewrite a prompt — it is that the strategies are named, documented and matched to prompt types, so "complex problem" points at tree_of_thoughts and "classification" at medprompt. Treat the improvement percentages in the README as the published research figures they are cited as, not as measurements of your prompt. Nothing here calls a model on your behalf, so there is no key to supply and no per-call cost.
A prompt-engineering server. It analyses a prompt, applies a chosen optimisation strategy or selects one automatically, and hands back production-ready templates for common professional tasks.
- analyze_prompt reports on a prompt's quality and names the issues it finds
- optimize_prompt applies a specific strategy — clarity, specificity, chain of thought, few-shot, structured output or role-based
- auto_optimize picks the strategy for you when you do not want to choose
- advanced_optimize applies the research-backed set: tree of thoughts, constitutional AI, automatic prompt engineer, meta-prompting, self-refine, TEXTGRAD, medprompt and PromptWizard
- get_prompt_template returns a basic template; get_domain_template returns a professional one and list_domain_templates browses what is available
- Templates cover eleven domains, including business analysis, product management, content creation, development, communication, strategy, operations, legal, customer experience, data analysis and meeting management
Python 3.8 or higher. Clone the repository, install from requirements.txt, and point the client at prompt_optimizer.py — the setup script setup_interactive.py writes the Claude Desktop configuration for you. No API key and no account; MCP_LOG_LEVEL set to debug turns on verbose logging.
One command — pip install -r requirements.txt