Claw Memory Bridge

Cross-agent shared memory bridge for MCP. Connects memory across QClaw, Claude Code, and Codex via MCP tools.

跨 Agent 共享记忆桥 — 通过 MCP 工具打通 QClaw、Claude Code、Codex 的记忆，让多个 AI agent 共享笔记与结论。

Features

• 存储与检索跨 agent 的笔记、决策、结论

• 按 agent 归档（claude-code / codex / qclaw / user）

• 全文搜索（SQLite FTS5，支持中文）

• 可选 TencentDB 集成（当存在 ~/.openclaw/state/memory-tdai/ 时自动读取）

Related MCP server: Project Memory MCP Server

MCP Tools

Install

git clone https://github.com/ncepuee/claw-memory-bridge.git
cd claw-memory-bridge
npm install

Requires Node.js >= 22 (uses built-in node:sqlite).

Configure

Claude Code (stdio)

claude mcp add claw-memory-bridge --scope user -- node /path/to/claw-memory-bridge/server.js

Claude Code Plugin (Marketplace, recommended)

claude plugin marketplace add https://github.com/ncepuee/claw-memory-bridge
claude plugin install claw-memory-bridge@claw-memory-bridge

Plugin install auto-runs npm install for dependencies. Don't enable both stdio and plugin at the same time (same MCP name).

Codex

codex mcp add claw-memory-bridge -- node /path/to/claw-memory-bridge/server.js

Claude Desktop / other MCP clients

{
  "mcpServers": {
    "claw-memory-bridge": {
      "command": "node",
      "args": ["/path/to/claw-memory-bridge/server.js"]
    }
  }
}

Data

• Local SQLite: ~/.claw-memory-bridge/shared.db

• Optional TencentDB: ~/.openclaw/state/memory-tdai/ (auto-detected; skipped if absent)

Usage

Agents call memory_context at the start of a task to pull recent + relevant context, and memory_store at the end to persist key conclusions.

memory_store({
  agent: "claude-code",                              // codex / qclaw / user
  topic: "decision",                                 // project / decision / conclusion / note
  content: "我们在 XXX 项目上决定用 YYY 方案，因为 ZZZ",
  source: "/path/to/file",
  tags: "project=XXX,framework=YYY"
})

License

MIT © ncepuee