WorkingMemory

Persistent personal memory for AI assistants — save, search, and recall across every MCP client.

我該用這個嗎

品質與安全性

A
說明品質
95%
結構描述完整度
100%
命名品質
95%
汙染風險
80%
權限相符程度
100%
協定合規性
100%

發現項目(3)

  • HIGHTool poisoning patterns detected
  • MEDIUMTool 'save_memory' description contains placeholder text在 save_memory 中
  • INFOTool description contains placeholder or incomplete text在 save_memory 中

根據工具定義與協定合規性的自動化分析。

上下文成本

~908Token(工具定義)
~2.0 KB典型回應大小
中等的注意力影響(128k 上下文的 0.71%)

這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高,可用於其他工作的注意力就越少。

安裝

一鍵安裝

將以下內容加入你的 `claude_desktop_config.json` 檔案:

{
  "mcpServers": {
    "memory": {
      "url": "https://app.workingmemory.ai/mcp"
    }
  }
}

遠端端點

https://app.workingmemory.ai/mcpstreamable-http

它能做什麼

工具清單

工具(4)

🟢 唯讀🟡 寫入🔴 刪除⚪ 未知
🟡save_memory(text, client)

Save a thought, insight, fact, or todo to the user's Working Memory. The text is parsed and stored as one or more structured memory items, searchable a moment later via search_memory. Also use this when the user pastes memories or notes brought from another assistant — pass the full pasted text; it is split into individual memories automatically.

輸入結構描述

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "description": "The thought or note to save."
    },
    "client": {
      "type": "object",
      "description": "Optional context about the originating client session. Helps users later find saves from a specific conversation. `conversation_id` groups multiple saves; `conversation_title` is human-readable.",
      "properties": {
        "conversation_id": {
          "type": "string",
          "description": "Opaque identifier; typically the client's conversation thread ID."
        },
        "conversation_title": {
          "type": "string",
          "description": "Human-readable conversation title (max 200 chars; longer values truncated server-side)."
        }
      },
      "additionalProperties": false
    }
  },
  "required": [
    "text"
  ]
}
🟢search_memory(query, k, conversation_id, dedupe_recent, client)

Search the user's Working Memory for relevant past notes and facts. Use this when the user references something they may have stored earlier, or to find related context before answering. Always include your conversation id in client.conversation_id.

輸入結構描述

{
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "description": "Semantic query."
    },
    "k": {
      "type": "number",
      "description": "Max results (1–20). Defaults to a sensible value."
    },
    "conversation_id": {
      "type": "string",
      "description": "Optional. Restrict results to notes stamped with this conversation_id at save_memory time. Empty / whitespace-only values are treated as absent."
    },
    "dedupe_recent": {
      "type": "boolean",
      "description": "Optional. When false, semantic matches that are also recent are still returned in `relevant` (raw semantic results rather than RAG-deduped). Defaults to true."
    },
    "client": {
      "type": "object",
      "description": "Context about the CURRENT client session. Pass your conversation/thread id as `conversation_id` — it identifies this session for cross-session memory instrumentation and does NOT filter results (use the top-level `conversation_id` parameter for filtering).",
      "properties": {
        "conversation_id": {
          "type": "string",
          "description": "Opaque identifier; typically the client's conversation thread ID."
        },
        "conversation_title": {
          "type": "string",
          "description": "Human-readable conversation title (max 200 chars; longer values truncated server-side)."
        }
      },
      "additionalProperties": false
    }
  },
  "required": [
    "query"
  ]
}
🔴manage_memory(action, id)

Manage an existing memory item. Currently supports deleting a memory by id (soft delete — recoverable for 30 days). Use search_memory to find the id first.

輸入結構描述

{
  "type": "object",
  "properties": {
    "action": {
      "type": "string",
      "enum": [
        "delete"
      ],
      "description": "The management action to perform."
    },
    "id": {
      "type": "string",
      "description": "The memory item id to act on."
    }
  },
  "required": [
    "action",
    "id"
  ]
}
⚪capture_memory(task_content, task_description, client)

Capture durable facts from a task you just completed (drafting an email, summarizing a document, making a decision), as a byproduct — the user does nothing. Pass the task output/content; durable facts are extracted and stored automatically. Always include your conversation id in client.conversation_id. Returns immediately; extraction happens in the background.

輸入結構描述

{
  "type": "object",
  "properties": {
    "task_content": {
      "type": "string",
      "description": "The task output/content to mine for durable facts (max 32KB)."
    },
    "task_description": {
      "type": "string",
      "description": "Optional one-line description of the task, as extraction framing."
    },
    "client": {
      "type": "object",
      "description": "Current client session context. Pass your conversation/thread id as conversation_id.",
      "properties": {
        "conversation_id": {
          "type": "string",
          "description": "Opaque conversation/thread id."
        },
        "conversation_title": {
          "type": "string",
          "description": "Human-readable title (max 200 chars)."
        }
      },
      "additionalProperties": false
    }
  },
  "required": [
    "task_content"
  ]
}

建議的提示詞

search_research
Search for information about [topic] using WorkingMemory
預期的工具: search_memory
find_specific
Find [specific item] using WorkingMemory
預期的工具: search_memory

社群

為此伺服器評分

證據

近期觀測

已驗證未記錄版本4 個工具
已驗證未記錄版本4 個工具