Agent-Native Calendar
Agent-Native Google Calendar - manage events, sync, and public booking
作者提供的说明(英文原文)
Task automation via SQLite, desktop Local API, or Mindwtr Cloud; read-only by default.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
MINDWTR_DB_PATH, MINDWTR_MCP_CLOUD_URL, MINDWTR_MCP_CLOUD_TOKEN, MINDWTR_MCP_API_URL, MINDWTR_MCP_API_TOKEN标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"mindwtr": {
"command": "npx",
"args": [
"-y",
"mindwtr-mcp"
],
"env": {
"MINDWTR_DB_PATH": "your-value",
"MINDWTR_MCP_CLOUD_URL": "your-value",
"MINDWTR_MCP_CLOUD_TOKEN": "your-value",
"MINDWTR_MCP_API_URL": "your-value",
"MINDWTR_MCP_API_TOKEN": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add mindwtr --env MINDWTR_DB_PATH=your-value --env MINDWTR_MCP_CLOUD_URL=your-value --env MINDWTR_MCP_CLOUD_TOKEN=your-value --env MINDWTR_MCP_API_URL=your-value --env MINDWTR_MCP_API_TOKEN=your-value -- npx -y mindwtr-mcp运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"mindwtr": {
"command": "npx",
"args": [
"-y",
"mindwtr-mcp"
],
"env": {
"MINDWTR_DB_PATH": "your-value",
"MINDWTR_MCP_CLOUD_URL": "your-value",
"MINDWTR_MCP_CLOUD_TOKEN": "your-value",
"MINDWTR_MCP_API_URL": "your-value",
"MINDWTR_MCP_API_TOKEN": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"mindwtr": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"mindwtr-mcp"
],
"env": {
"MINDWTR_DB_PATH": "your-value",
"MINDWTR_MCP_CLOUD_URL": "your-value",
"MINDWTR_MCP_CLOUD_TOKEN": "your-value",
"MINDWTR_MCP_API_URL": "your-value",
"MINDWTR_MCP_API_TOKEN": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
MINDWTR_DB_PATH envMINDWTR_MCP_CLOUD_URL envMINDWTR_MCP_API_URL env这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 dongdongbh 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:MINDWTR_MCP_CLOUD_TOKEN, MINDWTR_MCP_API_TOKEN
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(1.1.14)于3天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:Agent-Native Calendar, Excel MCP Server, MCP Server for Excel
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Agent-Native Google Calendar - manage events, sync, and public booking
Create and edit Excel files with AI: charts, PivotTables, formulas. No Excel needed.
Excel automation for AI - Sheets, Power Query, DAX, VBA, Tables, Ranges and more. Windows only.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
MCP server for Mindwtr. Connect MCP clients (Claude Desktop, etc.) to your local SQLite database, the desktop Local API, or a self-hosted Mindwtr Cloud endpoint.
By default this is a stdio server: MCP clients launch it as a subprocess and talk over JSON-RPC on stdin/stdout. It also has an opt-in HTTP transport (see Remote access (HTTP)) for self-hosters who want to expose it at a URL instead.
The desktop app includes a built-in MCP server in Settings → Integrations → Local MCP server (advanced). It requires no separately installed runtime or helper. The desktop installer does not add a standalone mindwtr-mcp command to your PATH; mobile apps do not host MCP.
You do not need to run the whole app from source to use MCP. You can use the normal desktop app binary for your tasks, then run this separate MCP helper from the repository with Bun, or build the helper once and run it with Node. Point the helper at the desktop app's local mindwtr.db, or use the Local API backend below.
On desktop, find the database in the default database locations listed under Requirements. Mobile binaries do not expose a local MCP server surface.
These requirements apply to the standalone helper, not the built-in desktop server.
--omit=optional. If no prebuilt binary is available for your platform, npm needs a working C/C++ build toolchain and Python for node-gyp.mindwtr.db) for SQLite mode, a running desktop Local API and bearer token for API mode, or a self-hosted Mindwtr Cloud URL and bearer token for Cloud modeFor SQLite mode with npm 12, approve the SQLite dependency's install script; otherwise a successful install can still fail at startup with a missing native binding. For npx or global installs, allow only better-sqlite3:
npx --allow-scripts=better-sqlite3 -y mindwtr-mcp --db "/path/to/mindwtr.db"
npm install -g --allow-scripts=better-sqlite3 mindwtr-mcp
In MCP client configurations using npx, add "--allow-scripts=better-sqlite3" before "mindwtr-mcp" in args. For an existing project-local install, run npm install-scripts approve better-sqlite3, then npm rebuild better-sqlite3 from that project. Review the dependency before approving; do not enable all dependency scripts.
Default database locations. The installed Windows and macOS apps keep the database in a data subfolder from version 1.3.2 on; Linux and the portable builds are unchanged. The server looks in both places, so it finds the database either way.
~/.local/share/mindwtr/mindwtr.db (Flatpak: ~/.var/app/tech.dongdongbh.mindwtr/data/mindwtr/mindwtr.db; Snap: ~/snap/mindwtr/current/.local/share/mindwtr/mindwtr.db)~/Library/Application Support/mindwtr/data/mindwtr.db~/Library/Application Support/mindwtr/mindwtr.db%APPDATA%\mindwtr\data\mindwtr.db%APPDATA%\mindwtr\mindwtr.dbAdditional macOS paths for sandboxed builds:
~/Library/Containers/tech.dongdongbh.mindwtr/Data/Library/Application Support/mindwtr/data/mindwtr.db~/Library/Containers/tech.dongdongbh.mindwtr/Data/Library/Application Support/mindwtr/mindwtr.dbA --db path pinned in an MCP client configuration keeps working across that move: if the file is not at the configured path, the server checks the matching data subfolder (or the parent folder) and reports on stderr which database it opened.
If mindwtr.db is missing but data.json exists in the same desktop data folder, the MCP server will bootstrap a fresh SQLite database from that local data snapshot on first start. The desktop data folder is the folder that holds mindwtr.db in the default locations above.
You can override local mode with:
--db /path/to/mindwtr.dbMINDWTR_DB_PATH=/path/to/mindwtr.dbMINDWTR_DB=/path/to/mindwtr.dbFor self-hosted Cloud mode, use:
--cloud-url https://mindwtr.example.com or MINDWTR_MCP_CLOUD_URL--cloud-token or MINDWTR_MCP_CLOUD_TOKEN--cloud-allow-insecure-http=true for trusted private HTTP deploymentsAfter installing the published package, run it directly:
mindwtr-mcp --db "/path/to/mindwtr.db"
Or let an MCP client launch it through npx:
{
"mcpServers": { "mindwtr": { "command": "npx", "args": [ "-y", "mindwtr-mcp", "--db", "~/.local/share/mindwtr/mindwtr.db" ] } } }
The npm package is read-only by default. Add --write only when you explicitly want add/update/complete/delete tools enabled.
Requires mindwtr-mcp 1.1.11 or later.
Enable the desktop app's Local API in Settings and copy its bearer token. Keep the app and API running. This backend uses the app's REST endpoints and never opens a database or falls back to SQLite.
For an API-only installation, omit the optional native SQLite addon:
npm install -g --omit=optional mindwtr-mcp
Configure your MCP client to pass the token through its environment, rather than putting it in command-line arguments:
{
"command": "mindwtr-mcp", "args": ["--api-url", "http://127.0.0.1:3456"], "env": { "MINDWTR_MCP_API_TOKEN": " " } }
Use the port shown in Settings. MINDWTR_MCP_API_URL can replace --api-url. URLs must use literal 127.0.0.1 or [::1], with no credentials, path, query or fragment; redirects are refused. Do not combine API and Cloud/--db options. Existing database-path environment variables are ignored in API mode.
Writes remain disabled unless explicitly enabled with --write. If the app is closed or the token is invalid, the helper returns an error. A timed-out write may have reached the app: inspect the task before retrying, as the helper does not retry writes automatically.
Desktop 1.3.8 adds optional protection for task creation and edits in Local API mode:
mindwtr_add_task accepts idempotencyKey, 1–128 visible ASCII characters without spaces. Keep the same key and identical input when retrying an uncertain create, including attachment links. The desktop retains the original result for seven days, even if that task is later edited or deleted. Different input with the same key returns idempotency_conflict. A full receipt store returns idempotency_capacity; retry later with the same key and input.mindwtr_get_task, task creation, and task edits return { "task": { ... }, "version": "sha256-..." }. Pass the version unchanged as expectedVersion to mindwtr_update_task. A stale edit returns task_version_conflict; read the task again and review the changes before retrying. An edit that changes nothing preserves the version.Both arguments are optional. Without them, writes keep their existing behavior. Protected requests use versioned desktop endpoints and fail on older desktop versions without falling back to an unprotected write. Older desktop reads can omit version. These arguments are unavailable in Cloud and direct SQLite mode; unsupported task input fields are rejected. Attachment replacement and lifecycle restrictions below still apply.
| MCP capability | Desktop Local API support |
|---|---|
| List/get tasks | Supported, including rich search operators, date filters, sorting and pagination |
| Create tasks | Explicit title and fields; quickAdd is unavailable |
| Update/complete/delete/restore tasks | Supported through the API's corresponding endpoints; terminal status changes cannot be submitted as ordinary edits |
| List/get/create/update/delete projects | Supported for the API's basic project fields: title, color, status, Area and sequential behavior |
| List areas | Supported |
| View section catalogue and project assignments | Unavailable; task assignments by known ID are supported |
| Sections, managed people, Area writes | Unavailable; these tools are hidden in API mode |
GTD availability view | Unavailable because the API does not expose the required section snapshot |
| Replacing attachment links | Unavailable because the API does not provide conditional attachment updates |
Other unsupported fields are rejected before a write. The desktop API's recurrence limitations also apply; some recurring completions must be performed in the app. This backend does not promise full SQLite/Cloud tool parity.
--api-url selects the helper's data source. --http instead selects how an MCP client connects to the helper; it can be used with any backend and has a separate authentication token.
Use Cloud mode when you run your own Mindwtr Cloud server and want MCP tools without pointing the helper at a local SQLite database:
npx -y mindwtr-mcp \
--cloud-url "https://mindwtr.example.com" \ --cloud-token "$MINDWTR_TOKEN"
Or pass the same values through environment variables:
MINDWTR_MCP_CLOUD_URL="https://mindwtr.example.com" \
MINDWTR_MCP_CLOUD_TOKEN="$MINDWTR_TOKEN" \ npx -y mindwtr-mcp
Cloud mode uses the self-hosted Cloud API. Reads come from the current /v1/data snapshot; with --write, task/project/section/area writes go through the Cloud server's per-resource REST endpoints (POST /v1/tasks, PATCH /v1/tasks/:id, and so on), so they get the same validation and revision stamping as any other client. Without --write, write tools return read_only. Person edits and restoring deleted tasks are not available in Cloud mode yet.
This does not make Mindwtr Cloud itself a hosted MCP server. It is still the same stdio helper, backed by a Cloud URL that you operate.
For private HTTP test deployments, local/private HTTP URLs are allowed by the shared Cloud client rules. Use --cloud-allow-insecure-http=true only for a self-hosted endpoint you intentionally trust.
By default mindwtr-mcp only speaks stdio. Pass --http to also (instead of stdio) serve a stateless streamable-HTTP MCP endpoint, so you can point a remote MCP client at a URL — the motivating case is Gemini Spark "custom apps", which take an MCP server URL. HTTP mode works with all three backends.
mindwtr-mcp --http --http-token "$(openssl rand -hex 32)" --db "/path/to/mindwtr.db"
Flags (all have MINDWTR_MCP_HTTP* env var equivalents):
--http / MINDWTR_MCP_HTTP — enable HTTP mode. Also implied by setting --http-host, --http-port, or --http-token.--http-token / MINDWTR_MCP_HTTP_TOKEN — required whenever HTTP mode is on, at least 16 characters. Generate one with openssl rand -hex 32. The server refuses to start without it — there is no way to expose HTTP mode unauthenticated, even on loopback.--http-host / MINDWTR_MCP_HTTP_HOST — bind address, default 127.0.0.1.--http-port / MINDWTR_MCP_HTTP_PORT — bind port, default 8722.The desktop app's built-in MCP server also uses port 8722. If it is on, pass another port with --http-port, for example --http-port 8723.
The MCP endpoint is POST /mcp and requires Authorization: Bearer on every request; GET /healthz returns 200 ok without auth for reverse-proxy health checks. Requests without a valid token get 401; repeated authentication failures get 429 with a Retry-After header. Bodies over 1 MiB get 413. When HTTP mode is on, the server does not also connect a stdio transport — it
只显示了开头部分;其余内容请在仓库中阅读