PromptTuner MCP

// Server instructions for MCP protocol // Based on 2024-2025 prompt engineering best practices export const SERVER_INSTRUCTIONS = `# PromptTuner MCP A professional prompt engineering toolkit that helps you write better prompts for AI assistants using modern best practices from Anthropic, OpenAI, and industry leaders. ## Quick Start | Goal | Tool | When to Use | |------|------|-------------| | Quick fix | \`refine_prompt\` | Fix typos, improve clarity | | Deep optimization | \`optimize_prompt\` | Apply multiple techniques | | Quality check | \`analyze_prompt\` | Get scores and suggestions | | Format detection | \`detect_format\` | Check Claude/GPT compatibility | ## Tools ### refine_prompt Improves a prompt using a single optimization technique. **Techniques:** - \`basic\` (default) - Fix grammar, spelling, and clarity - \`chainOfThought\` - Add step-by-step reasoning guidance - \`fewShot\` - Add input/output examples - \`roleBased\` - Add expert persona/role context - \`structured\` - Add XML (Claude) or Markdown (GPT) structure - \`comprehensive\` - Apply all techniques intelligently **Example:** \`\`\`json { "prompt": "help me code", "technique": "roleBased", "targetFormat": "claude" } \`\`\` ### analyze_prompt Scores your prompt (0-100) across five dimensions: - **Clarity** - Is the intent clear and unambiguous? - **Specificity** - Are requirements well-defined? - **Completeness** - Is context and output format specified? - **Structure** - Is it well-organized? - **Effectiveness** - Will it produce good results? Returns actionable suggestions for improvement. ### optimize_prompt Chains multiple techniques for maximum improvement. Shows before/after scores and a diff of changes. **Example:** \`\`\`json { "prompt": "write code", "techniques": ["basic", "roleBased", "structured"] } \`\`\` ### detect_format Identifies the target format (Claude XML, GPT Markdown, or JSON) with confidence score and recommendation. ## Target Formats | Format | Best For | Structure | |--------|----------|-----------| | \`claude\` | Anthropic Claude | XML tags: \`<context>\`, \`<task>\`, \`<requirements>\` | | \`gpt\` | OpenAI GPT | Markdown: \`## Context\`, \`## Task\`, \`**bold**\` | | \`json\` | Structured output | JSON schema specification | | \`auto\` | Auto-detect | Analyzes prompt to determine best format | ## Technique Selection Guide | Task Type | Recommended Techniques | |-----------|----------------------| | Simple query | \`basic\` only | | Code task | \`roleBased\` + \`structured\` | | Complex analysis | \`roleBased\` + \`chainOfThought\` | | Data extraction | \`structured\` + \`fewShot\` | | Creative writing | \`roleBased\` + \`fewShot\` | | Maximum quality | \`comprehensive\` | ## Resources - \`templates://catalog\` - Browse all template categories - \`templates://{category}/{name}\` - Get specific templates **Categories:** coding, writing, analysis, system-prompts, data-extraction ## Workflow Prompts - \`quick-optimize\` - Single-step optimization - \`full-analysis\` - Comprehensive review with scoring - \`convert-format\` - Transform to target format - \`best-practices-review\` - Educational feedback - \`iterative-improve\` - Multi-technique enhancement ## Modern Best Practices (2024-2025) ### Core Principles 1. **Be Specific** - Replace vague words ("something", "stuff", "etc.") with concrete terms 2. **Add Role Context** - "You are a senior engineer with expertise in..." activates domain knowledge 3. **Use Structure** - XML tags for Claude, Markdown for GPT (never mix) 4. **Show Examples** - 2-3 diverse examples demonstrate desired format 5. **Add Constraints** - Use ALWAYS/NEVER/MUST patterns for clear boundaries 6. **Specify Output** - Define exactly what format and structure you expect 7. **Enable Reasoning** - Use task-specific CoT triggers for complex tasks ### Advanced Techniques (from OpenAI & Anthropic Research) 8. **Place Instructions Twice** - For long prompts, put key instructions at BOTH beginning AND end 9. **Use Semantic Tags** - Claude: \`<context>\`, \`<task>\`, \`<requirements>\`, \`<output_format>\` 10. **Be Literal** - Modern models follow instructions more literally; be precise 11. **Add Quality Checks** - Include verification steps for important outputs 12. **Avoid Generic Roles** - "Helpful assistant" provides no benefit; use specific expert roles ### Prompt Architecture (Recommended Order) \`\`\` 1. Role/Identity (if applicable) 2. Context/Background 3. Task/Objective 4. Instructions/Steps 5. Requirements/Constraints (ALWAYS/NEVER) 6. Output Format 7. Examples (if helpful) 8. Final Reminder (reiterate critical instruction) \`\`\``;