Smart Prompts MCP Server

Instructions

Implementation Reference

• src/enhancedTools.ts:299-521 (handler)

  The handleHelp private method implements the core logic of the 'prompts_help' tool. It selects from predefined help texts based on the optional 'topic' parameter (general, workflow, creating, searching, github, examples) and returns the corresponding markdown-formatted guidance as a CallToolResult.
  private handleHelp(args: EnhancedToolArguments): CallToolResult { const topic = args.topic?.toLowerCase() || 'general'; const helpTexts: Record<string, string> = { general: `# 🚀 Smart Prompts MCP Server Help This server manages a library of reusable prompt templates stored in GitHub. ## 🎯 **RECOMMENDED WORKFLOW (Follow This Order):** ### 1. 🔍 **ALWAYS START WITH SEARCH** → Use **search_prompts** first to find existing prompts → Avoid creating duplicates and discover what's available ### 2. 📖 **GET DETAILS WHEN NEEDED** → Use **get_prompt** with exact names from search results → See full content and template variables ### 3. ✨ **CREATE ONLY WHEN NECESSARY** → Use **create_github_prompt** after confirming nothing similar exists → Automatically saves to GitHub with proper organization ### 4. 🔗 **COMBINE FOR COMPLEX TASKS** → Use **compose_prompts** to build multi-step workflows → Reuse existing prompts in new combinations ## 📚 **Available Tools:** 1. **prompts_help** - Get contextual help and guidance 2. **search_prompts** - 🔍 START HERE - Find existing prompts 3. **list_prompt_categories** - Browse all categories 4. **get_prompt** - Get full prompt details (use exact names from search) 5. **create_github_prompt** - Create new prompts (search first!) 6. **compose_prompts** - Combine multiple prompts 7. **check_github_status** - Verify GitHub connection 💡 **Pro Tip:** Search before you create! Most tasks already have prompts. Use 'prompts_help' with topic: "creating", "searching", "workflow", or "examples" for detailed guidance.`, workflow: `# 🔄 Recommended Workflow ## ✅ **STEP-BY-STEP PROCESS:** ### **DISCOVERY PHASE** 🔍 1. **Start with search_prompts** - Use keywords related to your task - Browse by category if unsure - Check multiple search terms 2. **Explore categories with list_prompt_categories** - See what's available in each area - Get counts to understand library size ### **EVALUATION PHASE** 📖 3. **Use get_prompt for promising results** - Get full content and metadata - Check template variables and examples - Evaluate if it meets your needs ### **ACTION PHASE** ⚡ 4a. **If perfect match found:** - Use the existing prompt directly - Note the template variables needed 4b. **If close match found:** - Consider using compose_prompts to combine - Extend with additional instructions 4c. **If no match found:** - Use create_github_prompt to build new one - Follow naming conventions - Add good metadata for future discovery ## 🎯 **EXAMPLE WORKFLOW:** \`\`\` User needs: "Help with code reviews" 1. search_prompts with query: "code review" → Finds: "Code Review Assistant" 2. get_prompt with name: "code_review_assistant" → Perfect! Has template variables for {{code}} and {{focus_areas}} 3. Use the existing prompt ✅ \`\`\` ## ❌ **ANTI-PATTERNS TO AVOID:** - Creating prompts without searching first - Using get_prompt with guessed names - Ignoring template variables and examples - Not checking categories for context`, creating: `# Creating Prompts ## Best Practices: 1. **Search first** - Check if a similar prompt exists 2. **Use clear names** - lowercase_with_underscores (e.g., api_documentation_generator) 3. **Categorize properly** - Use existing categories when possible 4. **Add good metadata** - Title, description, and tags help discoverability 5. **Include examples** - Show how to use the prompt in the content ## Example: \`\`\` create_github_prompt with: - name: "code_reviewer" - title: "Automated Code Review Assistant" - description: "Reviews code for best practices and potential issues" - category: "development" - content: "Review the following code for:\\n1. Best practices\\n2. Potential bugs\\n3. Performance issues\\n\\nCode:\\n{{code}}" - tags: ["code-review", "quality", "development"] - arguments: [{"name": "code", "description": "The code to review", "required": true}] \`\`\` ## Template Variables: Use {{variable_name}} in content for dynamic placeholders.`, searching: `# Searching for Prompts ## Search Methods: ### 1. Keyword Search \`\`\` search_prompts with: - query: "api documentation" \`\`\` Searches in titles, descriptions, and content. ### 2. Category Filter \`\`\` search_prompts with: - category: "development" \`\`\` Shows all prompts in a specific category. ### 3. Tag Filter \`\`\` search_prompts with: - tags: ["api", "rest"] \`\`\` Find prompts with specific tags. ### 4. Combined Search \`\`\` search_prompts with: - query: "test" - category: "development" - tags: ["automation"] \`\`\` ## Tips: - Start broad, then narrow down - Check list_prompt_categories first - Use partial keywords (e.g., "doc" finds "documentation")`, github: `# GitHub Integration ## How it Works: 1. Prompts are stored in GitHub repository: ${this.hasGitHubAccess ? 'Connected ✓' : 'Read-only (no write access)'} 2. Changes are committed directly to the repository 3. Other users see updates immediately ## Requirements: - GitHub token with repo access (currently ${this.hasGitHubAccess ? 'configured' : 'missing or read-only'}) - Repository: Check with check_github_status ## Creating Prompts: - Use create_github_prompt to save directly to GitHub - Prompts are organized in category folders - Each prompt is a markdown file with YAML frontmatter ## Troubleshooting: - Run check_github_status to verify connection - Ensure GITHUB_TOKEN environment variable is set - Token needs 'repo' scope for write access`, examples: `# Example Prompts ## 1. Simple Prompt \`\`\` create_github_prompt with: - name: "explain_code" - title: "Code Explanation Generator" - description: "Explains code in simple terms" - category: "development" - content: "Explain this code in simple terms:\\n\\n{{code}}" - tags: ["explanation", "documentation"] \`\`\` ## 2. Complex Prompt with Multiple Arguments \`\`\` create_github_prompt with: - name: "api_endpoint_generator" - title: "REST API Endpoint Generator" - description: "Creates REST endpoints with documentation" - category: "development" - content: "Create a {{method}} endpoint for {{resource}} that {{description}}.\\n\\nInclude:\\n- Validation\\n- Error handling\\n- OpenAPI docs" - arguments: [ {"name": "method", "description": "HTTP method", "required": true}, {"name": "resource", "description": "Resource name", "required": true}, {"name": "description", "description": "What the endpoint does", "required": true} ] \`\`\` ## 3. Composed Prompt \`\`\` compose_prompts with: - prompts: ["code_reviewer", "documentation_generator"] - separator: "\\n\\nNext Task:\\n\\n" \`\`\``, }; const helpText = helpTexts[topic] || helpTexts.general; return { content: [ { type: 'text', text: helpText, } as TextContent, ], }; }
• src/enhancedTools.ts:75-86 (schema)

  The tool definition in getToolDefinitions(), including name, description, and inputSchema specifying an optional 'topic' string parameter.
  name: 'prompts_help', description: 'Get help understanding how to use the Smart Prompts tools effectively. Returns guidance on tool usage, examples, and best practices.', inputSchema: { type: 'object', properties: { topic: { type: 'string', description: 'Specific topic to get help on (e.g., "creating", "searching", "github", "examples"). Leave empty for general help.', }, }, }, },
• src/index.ts:56-62 (registration)

  Registers the ListTools and CallTool request handlers on the MCP Server, delegating to EnhancedPromptTools.getToolDefinitions() (exposes prompts_help schema) and EnhancedPromptTools.handleToolCall() (executes prompts_help via internal switch). This effectively registers the tool with the MCP protocol.
  server.setRequestHandler(ListToolsRequestSchema, async () => { return enhancedTools.getToolDefinitions(); }); server.setRequestHandler(CallToolRequestSchema, async (request) => { return await enhancedTools.handleToolCall(request); });