Memory MCP Server

/** * MCP Prompts * 提供預定義的提示模板，幫助 AI 更好地使用記憶系統 */ /** * 創建提示處理器 */ export function createPrompts() { const prompts = [ { name: 'remember-user-info', description: 'Store important user information in long-term memory', arguments: [ { name: 'info_type', description: 'Type of information (e.g., preference, birthday, fact)', required: true }, { name: 'information', description: 'The information to remember', required: true }, { name: 'conversation_id', description: 'Conversation ID to store this memory under', required: false } ] }, { name: 'recall-context', description: 'Search for relevant memories based on current conversation', arguments: [ { name: 'current_topic', description: 'Current topic or question being discussed', required: true }, { name: 'conversation_id', description: 'Conversation ID to search within', required: false } ] }, { name: 'create-reminder', description: 'Create a conditional reminder that activates based on context or date', arguments: [ { name: 'reminder_content', description: 'What to remind about', required: true }, { name: 'trigger_condition', description: 'When to trigger (keywords or date)', required: true }, { name: 'conversation_id', description: 'Conversation ID for this reminder', required: false } ] }, { name: 'analyze-conversation', description: 'Analyze conversation history and create appropriate memories', arguments: [ { name: 'conversation_id', description: 'Conversation ID to analyze', required: true } ] } ]; /** * 處理提示獲取請求 */ async function getPrompt(name, args) { if (name === 'remember-user-info') { const { info_type, information, conversation_id = 'default' } = args || {}; return { messages: [ { role: 'user', content: { type: 'text', text: `Please store the following ${info_type || 'information'} in long-term memory: ${information || '[information not provided]'}` } }, { role: 'assistant', content: { type: 'text', text: `I'll help you store this information in long-term memory. Let me create a memory entry for this ${info_type || 'information'}. To properly store "${information || '[information]'}", I'll: 1. Create a descriptive name for the memory 2. Set up an appropriate trigger condition that will activate when relevant 3. Store it under conversation_id: "${conversation_id}" Here's my approach: **Memory Name**: ${info_type ? `${info_type.toLowerCase().replace(/\s+/g, '_')}_` : ''}${Date.now()} **Trigger Condition**: I'll use keywords from the information to create a trigger that activates when relevant topics are mentioned. **Recommended Trigger**: \`\`\`javascript match_keys(context.messages, [${information ? information.split(' ').slice(0, 3).map(w => `'${w}'`).join(', ') : "'keyword'"}], 'any') \`\`\` Now I'll use the add_long_term_memory tool to store this.` } } ] }; } if (name === 'recall-context') { const { current_topic, conversation_id = 'default' } = args || {}; return { messages: [ { role: 'user', content: { type: 'text', text: `Search for relevant memories about: ${current_topic || '[topic not provided]'}` } }, { role: 'assistant', content: { type: 'text', text: `I'll search for relevant memories about "${current_topic || '[topic]'}". I'll use both short-term and long-term memory searches: **Short-term Memory Search**: - Search recent conversation context - Look for related discussions - Find similar topics mentioned before **Long-term Memory Search**: - Check for permanent facts and preferences - Evaluate trigger conditions - Include random memories for serendipity Using conversation_id: "${conversation_id}" Let me search now using: 1. search_short_term_memories - for recent context 2. search_long_term_memories - for permanent facts The search will consider: - Keyword relevance - Time decay (recent memories are weighted higher) - Accumulated memory scores - Trigger condition matching` } } ] }; } if (name === 'create-reminder') { const { reminder_content, trigger_condition, conversation_id = 'default' } = args || {}; return { messages: [ { role: 'user', content: { type: 'text', text: `Create a reminder: ${reminder_content || '[content not provided]'} that triggers when: ${trigger_condition || '[condition not provided]'}` } }, { role: 'assistant', content: { type: 'text', text: `I'll create a conditional reminder for: "${reminder_content || '[content]'}" **Trigger Condition**: ${trigger_condition || '[condition]'} Let me convert this into a proper JavaScript trigger condition: ${trigger_condition && trigger_condition.toLowerCase().includes('date') ? ` **Date-based Trigger Example**: \`\`\`javascript // Triggers on a specific date or when mentioned (new Date().getMonth() === 6 && new Date().getDate() === 17) || match_keys(context.messages, ['reminder', 'important'], 'any') \`\`\` ` : ` **Keyword-based Trigger Example**: \`\`\`javascript // Triggers when specific keywords are mentioned match_keys(context.messages, [${trigger_condition ? trigger_condition.split(' ').slice(0, 3).map(w => `'${w}'`).join(', ') : "'keyword'"}], 'any') \`\`\` `} **Memory Details**: - Name: reminder_${Date.now()} - Prompt: ${reminder_content || '[content]'} - Conversation ID: ${conversation_id} I'll now use add_long_term_memory to create this reminder.` } } ] }; } if (name === 'analyze-conversation') { const { conversation_id = 'default' } = args || {}; return { messages: [ { role: 'user', content: { type: 'text', text: `Analyze conversation ${conversation_id} and suggest what should be remembered` } }, { role: 'assistant', content: { type: 'text', text: `I'll analyze conversation "${conversation_id}" and identify what should be remembered. **Analysis Steps**: 1. **Get Current Statistics** - Use get_memory_stats to see what's already stored - Check both short-term and long-term memory counts - Review memory scores and relevance 2. **Identify Important Information** - User preferences and personal information - Important facts and decisions - Recurring topics and themes - Action items and reminders 3. **Categorize for Storage** **Short-term Memory (temporary context)**: - Recent conversation flow - Current task context - Session-specific information **Long-term Memory (permanent facts)**: - Personal information (birthdays, preferences) - Important decisions - Persistent preferences - Scheduled reminders 4. **Create Memory Entries** - Use descriptive names - Set appropriate trigger conditions - Include context for future reference Let me first get the current memory statistics for this conversation using get_memory_stats.` } } ] }; } throw new Error(`Unknown prompt: ${name}`); } return { prompts, getPrompt }; } export default createPrompts;