splitwise-mcp
A local MCP server that exposes the Splitwise API as tools, enabling conversational management of expenses, groups, and friends. It runs entirely locally over stdio with no hosting required.
README
splitwise-mcp
A local MCP (Model Context Protocol) server that exposes the Splitwise API as tools, so MCP clients (Claude Desktop, Claude Code, etc.) can read and manage expenses, groups, and friends conversationally.
Runs entirely locally over stdio — no hosting, no public endpoint, no data leaves your machine except direct calls to secure.splitwise.com.
This is an independent, unofficial project. It is not affiliated with, endorsed by, or supported by Splitwise, Inc.
Requirements: Node.js 18+ and a Splitwise personal API key.
Tools
| Tool | Description |
|---|---|
get_current_user |
Identity of the authenticated account |
list_groups |
All groups with member balances |
get_group |
Full detail + simplified debts for one group |
list_friends |
All friends with balances owed/owing |
list_expenses |
Filterable expense list (group, friend, date range) |
get_expense |
Full detail for one expense |
create_expense |
Create an expense, split equally or with explicit shares |
create_expenses_bulk |
Create up to 50 expenses in one call; each succeeds/fails independently |
update_expense |
Update an existing expense |
delete_expense |
Delete an expense (destructive — clients should confirm first) |
delete_expenses_bulk |
Delete up to 100 expenses by id in one call (destructive — confirm first) |
get_balances |
Derived summary of who owes whom, across friends and groups |
Splitwise's API has no native bulk endpoint — the bulk tools fan out to the same per-expense calls as create_expense/delete_expense, with up to 5 requests in flight at once, and report a per-item ok/error result so a partial failure doesn't lose the rest of the batch.
Setup
- Get a Splitwise API key: go to secure.splitwise.com/apps, register an application, and copy the personal API key it gives you.
- Install dependencies and build:
npm install npm run build - Run the tests (optional, but recommended before wiring it into a client):
npm test
Client configuration
Add to your MCP client's config (e.g. Claude Desktop's claude_desktop_config.json, or Claude Code's .mcp.json):
{
"mcpServers": {
"splitwise": {
"command": "node",
"args": ["/absolute/path/to/splitwise-mcp/build/index.js"],
"env": {
"SPLITWISE_API_KEY": "your-personal-api-key"
}
}
}
}
Once published to npm, this can instead be:
{
"mcpServers": {
"splitwise": {
"command": "npx",
"args": ["-y", "splitwise-mcp"],
"env": { "SPLITWISE_API_KEY": "your-personal-api-key" }
}
}
}
Never commit your API key. Keep it only in your MCP client's config or a local, gitignored
.envfile — never in a tool input, a committed file, or a script argument that might end up in shell history.
Development
npm run dev # tsc --watch
npm test # vitest run
npm run test:watch # vitest
npm run format # prettier --write .
npm run format:check # prettier --check . (used in CI)
npm run inspector # build + launch MCP Inspector against the server
Design notes
- Money amounts are handled as decimal strings end-to-end (never parsed to float) — share validation for
create_expense/update_expenseuses integer-cent arithmetic. - Splitwise sometimes returns HTTP 200 with an
errorsobject in the body instead of a non-2xx status; the client checks the body, not just the status code. - Responses are trimmed of Splitwise's noisier fields (avatars, notification settings) before being returned to the model, to save context.
- See PLAN.md for the original design plan and v2 candidates (group/friend management, comments, MCP resources).
License
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。