add_long_term_memory

Store persistent memories that automatically activate when specific conversation conditions are met, using JavaScript triggers to determine relevance.

Instructions

Add a new long-term memory with a trigger condition. The trigger is JavaScript code that determines when this memory should be activated. Available context: context.messages (array), context.conversation_id (string), context.participants (object). Available functions: match_keys(messages, keywords, scope, depth), match_keys_all(messages, keywords, scope, depth).

Input Schema

TableJSON Schema

Name	Required	Description
`name`	Yes	Unique name for the memory
`prompt`	Yes	The memory content to be recalled when triggered
`trigger`	Yes	JavaScript code that returns true/false to determine if memory should activate. Example: "match_keys(context.messages, ['birthday'], 'any') \|\| new Date().getMonth() === 6"
`conversation_id`	No	Conversation ID that owns this memory (defaults to "default")
`createdContext`	No	Optional context about when/why this memory was created
`recentMessages`	No	Optional recent messages to auto-generate createdContext

Implementation Reference

src/tools/long-term-tools.js:32-66 (handler)

The main handler function for the 'add_long_term_memory' tool. It processes input arguments, auto-generates context if recentMessages provided, adds the memory via memoryManager.addMemory, saves to storage, and returns success/error response.

handler: async (args) => {
  try {
    // Auto-generate context if messages provided
    let createdContext = args.createdContext || '';
    if (!createdContext && args.recentMessages) {
      createdContext = createContextSnapshot(args.recentMessages, 4);
    }

    const result = await memoryManager.addMemory({
      name: args.name,
      prompt: args.prompt,
      trigger: args.trigger,
      createdContext
    });

    if (result.success) {
      await storageManager.saveLongTermMemories(memoryManager.getMemories());
      return {
        success: true,
        message: `Memory "${args.name}" added successfully`,
        totalMemories: memoryManager.getMemories().length
      };
    } else {
      return {
        success: false,
        error: result.error
      };
    }
  } catch (error) {
    return {
      success: false,
      error: error.message
    };
  }
}

src/tools/long-term-tools.js:21-31 (schema)

Zod inputSchema defining parameters for the tool: name, prompt, trigger, conversation_id (opt), createdContext (opt), recentMessages (opt).

inputSchema: z.object({
  name: z.string().describe('Unique name for the memory'),
  prompt: z.string().describe('The memory content to be recalled when triggered'),
  trigger: z.string().describe('JavaScript code that returns true/false to determine if memory should activate. Example: "match_keys(context.messages, [\'birthday\'], \'any\') || new Date().getMonth() === 6"'),
  conversation_id: z.string().optional().describe('Conversation ID that owns this memory (defaults to "default")'),
  createdContext: z.string().optional().describe('Optional context about when/why this memory was created'),
  recentMessages: z.array(z.object({
    role: z.enum(['user', 'assistant', 'system']),
    content: z.string()
  })).optional().describe('Optional recent messages to auto-generate createdContext')
}),

src/index.js:156-158 (registration)

Static registration of default long-term tools (including add_long_term_memory) to the toolRegistry for list_tools response.

// 注册所有长期记忆工具
const longTermTools = createLongTermTools(defaultLongTermManager, defaultStorageManager);
longTermTools.forEach(tool => registerTool(tool, 'long-term'));

src/index.js:291-295 (registration)

Dynamic registration/execution: recreates long-term tools with conversation-specific managers and calls the handler for 'add_long_term_memory' during tool invocation.

manager = await getLongTermManager(conversationId);
storage = getStorageManager(conversationId);
const tools = createLongTermTools(manager, storage);
const tool = tools.find(t => t.name === toolName);
result = await withTimeout(tool.handler(validatedArgs), timeout, `Tool ${toolName} timeout`);

Memory MCP Server