Supabase
MCP server for interacting with the Supabase platform
作者提供的说明(英文原文)
MCP server for dbatools — exposes SQL Server management commands as MCP tools
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。
DBATOOLS_SAFE_MODE, MAX_OUTPUT_ROWS, COMMAND_TIMEOUT_SECONDS, PWSH_EXE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": {
"dbatools": {
"command": "npx",
"args": [
"-y",
"dbatools-mcp-server"
],
"env": {
"DBATOOLS_SAFE_MODE": "your-value",
"MAX_OUTPUT_ROWS": "your-value",
"COMMAND_TIMEOUT_SECONDS": "your-value",
"PWSH_EXE": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add dbatools --env DBATOOLS_SAFE_MODE=your-value --env MAX_OUTPUT_ROWS=your-value --env COMMAND_TIMEOUT_SECONDS=your-value --env PWSH_EXE=your-value -- npx -y dbatools-mcp-server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": {
"dbatools": {
"command": "npx",
"args": [
"-y",
"dbatools-mcp-server"
],
"env": {
"DBATOOLS_SAFE_MODE": "your-value",
"MAX_OUTPUT_ROWS": "your-value",
"COMMAND_TIMEOUT_SECONDS": "your-value",
"PWSH_EXE": "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": {
"dbatools": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"dbatools-mcp-server"
],
"env": {
"DBATOOLS_SAFE_MODE": "your-value",
"MAX_OUTPUT_ROWS": "your-value",
"COMMAND_TIMEOUT_SECONDS": "your-value",
"PWSH_EXE": "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。
DBATOOLS_SAFE_MODE envMAX_OUTPUT_ROWS envCOMMAND_TIMEOUT_SECONDS envPWSH_EXE 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6日
注册名称已通过 GitHub 账户 dataplat 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。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
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(0.5.0)于7个月前进入注册中心。
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:Supabase, MongoDB MCP Server, Elasticsearch MCP Server
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
MCP server for interacting with the Supabase platform
MongoDB Model Context Protocol Server
MCP server for interacting with Elasticsearch
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。
A Model Context Protocol (MCP) server for the dbatools PowerShell module.
Exposes dbatools commands as MCP tools so AI assistants (GitHub Copilot, Claude, etc.) can discover, explain, and execute dbatools commands directly — with all metadata sourced from dbatools' own comment-based help.
Get-Help -FullreadOnlyHint / destructiveHint / idempotentHint / openWorldHint so MCP clients (e.g. VS Code) can surface risk in the UI before a tool is even calleddbatools://command/{name} to read any command's full normalized help as structured JSON, independent of the chat-oriented tool responsesinvoke_dbatools_command reuses one long-lived pwsh process with dbatools pre-imported instead of spawning (and re-importing dbatools into) a fresh process on every callconfirm: true to executeSqlCredential: { username, password } for SQL auth instancespwsh)Install-Module dbatools -Scope CurrentUser
# 1. Clone the repo
git clone https://github.com/Dataplat/dbatools-mcp-server.git cd dbatools-mcp-server
npm install
npm run refresh-help
npm run build
Then open the folder in VS Code — the .vscode/mcp.json file automatically registers the MCP server.
The included .vscode/mcp.json registers the server as a local STDIO MCP server. Open this folder in VS Code and the server will appear in the GitHub Copilot MCP panel.
{
"servers": { "dbatools": { "type": "stdio", "command": "node", "args": ["${workspaceFolder}/dist/server.js"], "env": { "DBATOOLS_SAFE_MODE": "true", "MAX_OUTPUT_ROWS": "100", "COMMAND_TIMEOUT_SECONDS": "60" } } } }
All settings are controlled via environment variables (set in .vscode/mcp.json or your shell):
| Variable | Default | Description |
|---|---|---|
PWSH_EXE | pwsh | Path to PowerShell executable |
DBATOOLS_SAFE_MODE | true | When true, non-readonly commands require confirm: true |
MAX_OUTPUT_ROWS | 100 | Maximum rows returned per command execution |
COMMAND_TIMEOUT_SECONDS | 60 | Seconds before PowerShell process is killed |
DBATOOLS_PERSISTENT_SESSION | true | When true, invoke_dbatools_command reuses one long-lived PowerShell process (dbatools pre-imported) instead of spawning a new process per call. Set to false to force a fresh process per call. |
The help index (generated/dbatools-help.json) is generated from your locally installed dbatools module. Re-run whenever dbatools is updated:
Update-Module dbatools -Scope CurrentUser
npm run refresh-help
The server detects version mismatches at runtime and warns you when the index is stale.
Commands are automatically classified by verb:
| Risk Level | Verbs | Behavior |
|---|---|---|
readonly | Get, Test, Find, Compare, … | Always allowed |
change | Set, New, Add, Copy, Enable, … | Requires confirm: true in safe mode |
destructive | Remove, Drop, Disable, Reset, … | Requires confirm: true in safe mode |
Each risk tier maps to standard MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint), so compatible clients can surface risk in their UI before a tool is even invoked. invoke_dbatools_command is a generic dispatcher that can run any command, so its annotation conservatively reflects the worst case (destructiveHint: true) — the actual per-command risk is still enforced at runtime by the safe-mode gate.
Besides the four tools, the server exposes every indexed command as an MCP resource at dbatools://command/{name} (e.g. dbatools://command/Get-DbaDatabase), returning the same normalized help record as structured JSON — useful for clients that browse/ground on resources rather than calling a tool. The browsable resource list excludes PowerShell aliases (e.g. Attach-DbaDatabase, an alias for Mount-DbaDatabase) since they just forward to another command's help.
By default, invoke_dbatools_command reuses a single long-lived pwsh process with dbatools pre-imported, instead of spawning a fresh process (and re-importing dbatools) on every call. This removes the ~1-2s+ process-startup and module-import cost from every invocation after the first.
A couple of things to be aware of:
Connect-DbaInstance) within a PowerShell session. With a persistent session, that caching is now visible across tool calls within one MCP server run, where previously every call was fully isolated. Each individual script still runs in its own scope, so plain variables never leak between calls — only dbatools' own internal caching behavior is affected.Set DBATOOLS_PERSISTENT_SESSION=false to disable this and go back to one fresh process per call.
For SQL-auth-only instances (e.g. Docker), pass credentials via the SqlCredential parameter:
{
"SqlInstance": "localhost,1433", "SqlCredential": { "username": " ", "password": "YourPassword" } }
dbatools-mcp-server/
├── src/ │ ├── server.ts # MCP server entry point, tool/resource definitions │ ├── powershell.ts # One-shot PowerShell process runner, health checks, version detection │ ├── powershell-session.ts # Persistent PowerShell session (long-lived process, dbatools pre-imported) │ ├── help-indexer.ts # Help manifest loader and command search │ ├── tool-registry.ts # Risk classification, tool annotations, safe argument builder │ └── types.ts # Shared TypeScript interfaces ├── scripts/ │ └── refresh-help.ps1 # Generates generated/dbatools-help.json ├── generated/ # Help index (gitignored, generated locally) ├── .vscode/ │ └── mcp.json # VS Code MCP local server registration └── dist/ # Compiled output (gitignored)
Contributions are welcome! Please open an issue first for significant changes.
This project follows the same community spirit as dbatools.
MIT — © 2026 DataPlat contributors