PromptArchitect MCP

Instructions

Implementation Reference

• src/tools/refinePrompt.ts:27-92 (handler)

  The core handler function that implements the refine_prompt tool logic. It calls the PromptArchitect API if available, otherwise falls back to basic refinements, and computes metadata.
  export async function refinePrompt(input: RefinePromptInput): Promise<{ refinedPrompt: string; changes: string[]; metadata: { originalWordCount: number; refinedWordCount: number; structurePreserved: boolean; }; }> { const { prompt, feedback, preserveStructure = true, targetModel, workspaceContext } = input; logger.info('Refining prompt', { promptLength: prompt.length, feedbackLength: feedback.length, preserveStructure, targetModel, hasWorkspaceContext: !!workspaceContext }); let refinedPrompt: string = ''; let changes: string[] = []; // Use PromptArchitect API if (isApiClientAvailable()) { try { const response = await apiRefinePrompt({ prompt, feedback, preserveStructure, targetModel, workspaceContext, }); refinedPrompt = response.refinedPrompt; changes = response.changes || ['Prompt refined based on feedback']; logger.info('Refined via PromptArchitect API'); } catch (error) { logger.warn('API request failed, using fallback', { error: error instanceof Error ? error.message : 'Unknown error' }); } } // Fallback basic refinement if (!refinedPrompt) { refinedPrompt = applyBasicRefinements(prompt, feedback); changes = ['Applied basic refinements']; logger.warn('Using fallback refinement'); } // Calculate metadata const originalWordCount = prompt.split(/\s+/).length; const refinedWordCount = refinedPrompt.split(/\s+/).length; const originalHasStructure = /^#+\s|^\d+\.|^-\s|^\*\s/m.test(prompt); const refinedHasStructure = /^#+\s|^\d+\.|^-\s|^\*\s/m.test(refinedPrompt); const structurePreserved = !originalHasStructure || refinedHasStructure; return { refinedPrompt, changes, metadata: { originalWordCount, refinedWordCount, structurePreserved, }, }; }
• src/tools/refinePrompt.ts:14-23 (schema)

  Zod schema defining the input validation for the refine_prompt tool.
  export const refinePromptSchema = z.object({ prompt: z.string().min(1).describe('The current prompt to refine'), feedback: z.string().min(1).describe('What should be improved or changed'), preserveStructure: z.boolean().optional().default(true).describe('Whether to preserve the original structure'), targetModel: z.enum(['gpt-4', 'claude', 'gemini', 'general']) .optional() .default('general') .describe('Target AI model to optimize for'), workspaceContext: z.string().optional().describe('Project context including file structure, tech stack, dependencies, and any relevant code snippets to ensure the refined prompt aligns with the project scope'), });
• src/server.ts:114-154 (registration)

  MCP tool list registration for 'refine_prompt' including name, description, and input schema.
  { name: 'refine_prompt', description: `Iteratively improve an existing prompt based on specific feedback. Use this tool when you need to: • Improve a prompt that didn't get good results • Add missing context or constraints • Make a prompt more specific or clearer • Adapt a prompt for a different AI model The tool preserves the original structure while applying targeted improvements. IMPORTANT: When available, pass workspace context (file structure, package.json, tech stack) to ensure refined prompts comply with the user's project scope and original request.`, inputSchema: { type: 'object', properties: { prompt: { type: 'string', description: 'The current prompt to refine.', }, feedback: { type: 'string', description: 'What should be improved. Examples: "make it more specific", "add error handling requirements", "focus on performance".', }, preserveStructure: { type: 'boolean', description: 'Whether to keep the original structure. Default: true.', }, targetModel: { type: 'string', enum: ['gpt-4', 'claude', 'gemini', 'general'], description: 'Target AI model for optimization.', }, workspaceContext: { type: 'string', description: 'Project context to ensure the refined prompt aligns with the codebase. Include: file/folder structure, package.json dependencies, tech stack (React, Node, etc.), relevant code snippets, and the original user request. This ensures the refined prompt complies with project conventions and scope.', }, }, required: ['prompt', 'feedback'], }, },
• src/server.ts:220-231 (registration)

  MCP tool call handler for 'refine_prompt' that parses input using the schema and invokes the refinePrompt handler.
  case 'refine_prompt': { const input = refinePromptSchema.parse(args); const result = await refinePrompt(input); return { content: [ { type: 'text', text: JSON.stringify(result, null, 2), }, ], }; }
• src/tools/refinePrompt.ts:94-120 (helper)

  Fallback helper function for basic prompt refinements when API is unavailable.
  function applyBasicRefinements(prompt: string, feedback: string): string { let refined = prompt; const feedbackLower = feedback.toLowerCase(); // Basic transformations based on common feedback if (feedbackLower.includes('more specific') || feedbackLower.includes('more detail')) { refined += '\n\n## Additional Requirements\n- Provide specific, detailed information\n- Include concrete examples where applicable'; } if (feedbackLower.includes('shorter') || feedbackLower.includes('concise')) { // Try to simplify refined = refined.replace(/\n\n+/g, '\n\n'); } if (feedbackLower.includes('structure') || feedbackLower.includes('organize')) { if (!/^#+\s/m.test(refined)) { refined = '## Task\n' + refined + '\n\n## Output\nProvide a well-structured response.'; } } if (feedbackLower.includes('example')) { refined += '\n\n## Example\nProvide a relevant example in your response.'; } return refined; }