Skip to main content
Glama
sqlite-memory-mcp

sqlite-memory-mcp

by miles990

sqlite-memory-mcp

npm version License: MIT

統一的 SQLite Memory MCP Server,為 Claude Code 生態系提供智能記憶管理

特色

  • 跨專案記憶共享 — 學一次,處處可用

  • FTS5 全文搜尋 — 毫秒級搜尋,精確匹配

  • Skill 效果追蹤 — 知道什麼最有效

  • 失敗經驗索引 — 不重複犯錯

  • Context 狀態共享 — 跨 Skill 無縫傳遞

  • 零外部依賴 — 純 SQLite,無需 PyTorch/ONNX

Related MCP server: Memory MCP

效能

指標

傳統方案

sqlite-memory-mcp

Token/搜尋

~2300

~200 (-91%)

搜尋速度

~20ms

~3.5ms (5.7x)

外部依賴

PyTorch/ONNX

無

並發支援

JSONL 無

SQLite WAL

安裝

從 npm 安裝(推薦)

npm install -g sqlite-memory-mcp

從源碼安裝

git clone https://github.com/miles990/claude-memory-mcp.git
cd claude-memory-mcp
npm install
npm run build

配置 Claude Code

在 ~/.claude/.mcp.json 加入:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["sqlite-memory-mcp"]
    }
  }
}

或如果從源碼安裝:

{
  "mcpServers": {
    "memory": {
      "command": "node",
      "args": ["/path/to/claude-memory-mcp/dist/index.js"]
    }
  }
}

工具列表 (23 tools)

Memory 工具 (6)

工具

說明

memory_write

寫入記憶到知識庫

memory_read

讀取特定記憶

memory_search

FTS5 全文搜尋

memory_list

列出記憶(可過濾)

memory_delete

刪除記憶

memory_stats

統計資訊

Skill 工具 (7)

工具

說明

skill_register

註冊 skill 安裝

skill_get

取得 skill 資訊

skill_list

列出所有 skill

skill_usage_start

開始使用追蹤

skill_usage_end

結束使用追蹤

skill_recommend

智能推薦(基於成功率)

skill_stats

使用統計

Context 工具 (5)

工具

說明

context_set

設定 context 值

context_get

取得 context 值

context_list

列出 session context

context_clear

清除 context

context_share

跨 session 共享

Failure 工具 (5)

工具

說明

failure_record

記錄失敗經驗

failure_search

FTS5 搜尋解法

failure_list

列出失敗記錄

failure_update

更新解法

failure_stats

失敗統計

資料庫

位置:~/.claude/claude.db

自動建立 schema,包含:

  • memory 表 + memory_fts FTS5

  • skills 表

  • skill_usage 表

  • failures 表 + failures_fts FTS5

  • context 表

使用範例

記憶搜尋

memory_search query="TypeScript pattern"
memory_list scope="global"

Skill 追蹤

skill_usage_start skill_name="evolve"
skill_usage_end usage_id=1 success=true outcome="completed"
skill_recommend project_type="typescript"

失敗經驗

failure_record error_pattern="TypeError: undefined" solution="Check null values"
failure_search query="TypeError undefined"

與競品比較

功能

server-memory

doobidoo

sqlite-memory-mcp

存儲

JSONL

SQLite-vec

SQLite WAL

搜尋

關鍵字

向量

FTS5 全文

Skill 追蹤

-

-

有

失敗索引

-

-

有

Context 共享

-

-

有

外部依賴

無

PyTorch

無

與 evolve skill 整合

此 MCP Server 設計為與 self-evolving-agent 整合:

  • CP1: 使用 memory_search + failure_search 搜尋經驗

  • CP3.5: 使用 memory_write 記錄學習

  • CP5: 使用 failure_record 記錄失敗

  • Skill 追蹤: 自動追蹤使用成功率

License

MIT

Available Tools

23 tools
context_clearC

Clear context values

Inputs: session_id.

context_getC

Get a context value

Inputs: key (required), session_id (required).

context_listA

List all context values for a session

Inputs: session_id (required).

context_setB

Set a context value for cross-skill state sharing

Inputs: key (required), value (required), session_id (required), skill_name, +1 more.

context_shareC

Share context from one session to another

Inputs: to_session (required), from_session (required), keys.

failure_listC

List failure experiences

Inputs: limit, skill_name.

failure_recordC

Record a failure experience for future reference

Inputs: error_pattern (required), solution, skill_name, project_path, +1 more.

failure_statsC

Get failure statistics

Inputs: no named parameters.

failure_updateB

Update a failure record with a solution

Inputs: id (required), solution (required).

memory_deleteB

Delete a memory entry

Inputs: key (required).

memory_listC

List memory entries with optional filters

Inputs: limit, scope, prefix.

memory_readA

Read a specific memory entry by key

Inputs: key (required).

memory_statsC

Get memory statistics

Inputs: no named parameters.

memory_writeC

Write a memory entry to the knowledge base

Inputs: key (required), content (required), tags, scope, +1 more.

skill_getC

Get skill information

Inputs: name (required).

skill_listA

List all registered skills

Inputs: project_path.

skill_recommendC

Get skill recommendations based on project type and success rates

Inputs: limit, project_type.

skill_registerC

Register a skill installation

Inputs: name (required), source (required), version (required), installed_by, +1 more.

skill_statsB

Get skill usage statistics

Inputs: no named parameters.

skill_usage_endC

End skill usage tracking

Inputs: success (required), usage_id (required), notes, outcome, +1 more.

skill_usage_startC

Start tracking skill usage

Inputs: skill_name (required), project_path.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.1/5.0

    Scored across 23 tools

    Disambiguation4/5

    Tools are grouped into clear domains: context, memory, skill, and failure, which makes their purposes generally obvious. A few retrieval tools like memory_read, memory_list, and memory_search have overlapping connotations, but their descriptions specify distinct lookup modes.

    Naming Consistency5/5

    All tool names follow a consistent lowercase snake_case verb_noun pattern, such as memory_write, context_set, skill_register, and failure_record. Each domain uses predictable verbs, so the naming convention is uniform and scannable.

    Tool Count4/5

    23 tools is on the higher end and feels slightly heavy, but the server covers four distinct subdomains that each need CRUD, search, and stats operations. The count is reasonable given the combined scope, though it could be tightened.

    Completeness4/5

    Each core domain has solid coverage: create, read, list, search, and delete/update where expected. Minor gaps exist, such as no memory update, no single-key context delete, and no skill unregister, but agents can generally work around these.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Provides persistent cross-session memory and full-text search for AI coding assistants, storing project context, decisions, and preferences while enabling searchable access to conversation history via local SQLite.
      8
      1
      MIT
    • -
      license
      Not graded
      quality
      D
      maintenance
      Provides persistent memory for AI assistants like Claude, storing and retrieving information across conversations using a local SQLite database.
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables Claude Code to query and store memories from past conversations using FTS5 search and topic-based retrieval, with no API costs.
      54 npm
      1
      Apache 2.0