handoff-mcp
Shared memory hub for LLMs to persist and share project context, enabling seamless handoffs between different AI agents.
README
handoff-mcp

MCP server that acts as a shared memory hub across multiple LLMs (Claude, Gemini, Copilot, Codex). Any agent entering a project immediately understands its architecture, patterns, and decisions — and can pick up exactly where the previous LLM left off.
Tools
| Tool | Description |
|---|---|
create_or_get_project |
Initialize or retrieve a project |
save_architecture |
Save architecture overview and tech stack |
get_project_summary |
Full project overview — perfect as first call |
save_context_snapshot |
Save working state before handing off |
get_context_snapshot |
Retrieve what the previous LLM was doing |
list_sessions |
List all saved sessions |
save_code_pattern |
Save reusable code patterns (JWT, FCM, etc.) |
get_code_patterns |
Retrieve patterns by category/language |
save_architectural_decision |
Save an ADR (why a decision was made) |
get_decisions |
Retrieve architectural decisions |
search_context |
Search across patterns, decisions, and snapshots |
Setup
No install needed. Just add the config below to your tool and it runs automatically via bunx.
Requires Bun installed on your machine. Install it with:
curl -fsSL https://bun.sh/install | bash
Claude Desktop
File: ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"handoff-mcp": {
"command": "bunx",
"args": ["handoff-mcp"]
}
}
}
Claude Code
claude mcp add handoff-mcp bunx handoff-mcp
Gemini CLI
File: ~/.gemini/settings.json
{
"mcpServers": {
"handoff-mcp": {
"command": "bunx",
"args": ["handoff-mcp"]
}
}
}
GitHub Copilot (VS Code)
File: .vscode/mcp.json in your workspace
{
"servers": {
"handoff-mcp": {
"type": "stdio",
"command": "bunx",
"args": ["handoff-mcp"]
}
}
}
OpenAI Codex CLI
File: ~/.codex/config.yaml
mcpServers:
handoff-mcp:
command: bunx
args:
- handoff-mcp
Shared database (recommended)
By default each tool creates its own context.db in the working directory. To share the same memory across all LLMs, point them all to the same file:
{
"mcpServers": {
"handoff-mcp": {
"command": "bunx",
"args": ["handoff-mcp"],
"env": { "DB_PATH": "/Users/you/handoff.db" }
}
}
}
Flow
Claude works on the project
→ create_or_get_project("my-app")
→ save_architecture({ description: "React Native + Spring Boot" })
Claude runs low on tokens
→ save_context_snapshot({
llmModel: "claude-opus-4",
taskDescription: "Implementing push notifications",
recentChanges: "FCM handler done",
nextSteps: "Wire up Foreground Service, test Android 16"
})
Gemini takes over
→ get_project_summary("my-app") # full context in one call
→ get_context_snapshot("my-app") # picks up exactly where Claude stopped
→ continues...
Dev
git clone https://github.com/Juan-Severiano/handoff-mcp
cd handoff-mcp
bun install
bun run dev # watch mode
bun run inspect # MCP Inspector
bun run build # compile to dist/
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。