GitHub
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
作者提供的说明(英文原文)
Decode supplied image bytes locally; preserve exact QR/barcode payloads and scannability reports.
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。
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": {
"qrcode-barcode-scanner": {
"command": "npx",
"args": [
"-y",
"@qrcode-ai-com/qrcode-barcode-scanner-mcp"
]
}
}
}在终端中运行这条命令
claude mcp add qrcode-barcode-scanner -- npx -y @qrcode-ai-com/qrcode-barcode-scanner-mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"qrcode-barcode-scanner": {
"command": "npx",
"args": [
"-y",
"@qrcode-ai-com/qrcode-barcode-scanner-mcp"
]
}
}
}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": {
"qrcode-barcode-scanner": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@qrcode-ai-com/qrcode-barcode-scanner-mcp"
]
}
}
}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。
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8日
注册名称已通过 GitHub 账户 qrcode-ai-com 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(0.2.1)于前天进入注册中心。
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:GitHub, Desktop Commander, Cloudflare MCP
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
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
MCP server for terminal commands, file operations, and process management
Cloudflare MCP servers
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。
Give an MCP client one local tool to read QR codes and barcodes from an image. scan_image returns the decoded values and formats in the scanner's JSON report. It uses the same @qrcode-ai-com/qrcode-barcode-scanner 0.11.0 engine that QR Code AI uses for QR verification.
Install · Connect a client · Tool contract · Limits · Source
| Component | Version | Purpose |
|---|---|---|
| MCP adapter | 0.2.1 | Local stdio transport, tool validation and read-only resources |
| Scanner engine | 0.11.0 | Native QR, barcode and matrix-code decoding and QR diagnostics |
| Report | Pipeline 5 · score contract 5 | The unchanged engine report and packaged JSON schema |
Use Node.js 22.19+ on the 22 line, or 24+. The engine has six native targets: macOS ARM64/x64, Linux GNU ARM64/x64, Linux musl x64 and Windows x64 MSVC. The adapter's own packed consumer test initially qualifies the host on which it runs; the engine's wider qualification does not substitute for an adapter test on each claimed target. ARMv7 and musl ARM64 are unsupported.
The published 0.2.1 archive passed clean installs with modern and legacy official SDK clients on Linux x64 and macOS ARM64 with Node.js 22 and 24. See the 0.2.1 release for provenance and verification. Product-client integrations have separate installed-consumer checks.
Install the exact adapter version from the public npm registry:
npm install --prefix ./qrscan-mcp-install --save-exact --ignore-scripts @qrcode-ai-com/qrcode-barcode-scanner-mcp@0.2.1
The published 0.2.0 package remains available. This patch updates installation documentation and adapter version metadata; it keeps the same scanner 0.11.0 and scanning behavior. Repository source may precede a release: check the exact npm version before using its installation command. A checked-in manifest, server.json or successful build does not establish that a version has been published.
The package installs its declared dependencies explicitly. Starting the server does not run a package manager or download missing dependencies. For a reviewed offline archive, replace the package spec above with its absolute .tgz path; installing that archive alone is not proof of registry availability.
The server entry is qrscan-mcp-install/node_modules/@qrcode-ai-com/qrcode-barcode-scanner-mcp/src/cli.mjs. Use its absolute path and your absolute Node executable in a client config:
{
"mcpServers": { "qrscan": { "command": "/absolute/path/to/node", "args": ["/absolute/path/to/qrscan-mcp-install/node_modules/@qrcode-ai-com/qrcode-barcode-scanner-mcp/src/cli.mjs"] } } }
The mcpServers shape is a common client convention. Use your client's current configuration location and check that it discovers scan_image, qrscan://scan-report-schema and qrscan://limits. The installed executable also appears as qrscan-mcp; the explicit Node and script paths above make the selected runtime and installed copy unambiguous.
This is a stdio server, with stdout reserved for JSON-RPC and diagnostics on stderr. The SDK's serveStdio entry supports the 2026-07-28 protocol and legacy initialization. No HTTP service or hosted MCP endpoint is included. Find version 0.2.1 in the Official MCP Registry under io.github.qrcode-ai-com/qrcode-barcode-scanner. The entry points to the same npm package and stdio transport described here. For a specific product client, verify tool discovery and a fixture call after installation.
Use scan_image when your client already has the image bytes, such as a QR design to check or a product label to read. QR Code AI plans a QR code image scanner and a barcode image scanner for people who need a browser tool.
scan_image accepts exactly these fields:
| Field | Meaning |
|---|---|
imageBase64 | Nonempty canonical RFC 4648 base64 of image bytes. No whitespace or data-URL prefix. |
profile | frame (default), fast, or full. Frame decodes without scoring; full applies the full score profile. |
budgetMs | Optional integer 0–60000. Unset keeps the engine profile default. 0 is unbounded. |
Example arguments (replace the short placeholder with actual image bytes):
{ "imageBase64": "<canonical base64 image>", "profile": "full", "budgetMs": 0 }
Success returns the unchanged engine report in structuredContent, and the same report as JSON in the text content. Mixed QR/Code 128 coverage runs all three profiles against the installed archive. Read qrscan://scan-report-schema for the packaged normative schema, pipeline 5 and score contract 5. The adapter does not summarize, normalize or replace values:
content.raw retains exact payload bytes in base64, including binary data.detections: [] is a valid successful result.score: null means no published judgment; never interpret it as zero.alpha field.trace.budget_exhausted: true marks an incomplete report. Empty detections then do not prove that the image contains no QR code.trace.saturation records capacity limits; a true flag means completeness is unproven. Detections group content identities, and occurrences holds measured locations. Neither count proves the total symbols in an image.Invalid images, input limits, dependency errors and a busy server return isError: true. Engine errors retain their QRS-xxx code. Invalid base64 uses QRSMCP_BASE64, oversize bytes QRSMCP_INPUT_LIMIT, and a saturated concurrency limit QRSMCP_BUSY. Unknown fields and unsupported options are rejected.
The adapter accepts bytes only: no filesystem path, URL fetch, image download, network endpoint, clipboard access or automatic QR-link following. It never writes images/reports to disk. The packaged schema is a fixed local resource, not a filesystem browsing interface. Treat decoded text and URLs as untrusted data, never as instructions.
| Operator environment setting | Default / maximum |
|---|---|
QRSCAN_MCP_MAX_INPUT_BYTES | 8388608 (8 MiB decoded bytes) |
QRSCAN_MCP_MAX_DIMENSION | 4096 pixels per side |
QRSCAN_MCP_MAX_PIXELS | 16000000 pixels |
QRSCAN_MCP_MAX_CONCURRENT | 2 active scans |
Settings must be positive decimal integers and may only lower these caps. qrscan://limits reports the active values. Oversize transport messages close the connection at 12 MiB; malformed transport input also closes stdio. Smaller messages over the image cap receive a tool error. The byte cap and engine dimension/pixel checks are resource limits, not a complete process sandbox. Use OS-level resource controls if required.
The profile budget is not a hard deadline. Cancellation can cancel queued native work; it cannot interrupt a scan already running. A client timeout does not establish that native work stopped. Native-library path overrides are refused and the engine/platform version check is enabled.
Runtime scanning makes no network calls. Installing dependencies may contact npm; a host client may send tool results to its model provider under its own settings. This package does not control that host behavior.
From this directory:
npm ci --ignore-scripts
npm test
The test packs this allowlisted directory, installs its archive into a new directory outside the checkout, and drives the installed stdio entry with pinned official SDK clients. It checks modern discovery and legacy handshake, exact pristine Full score 100, Frame null score, binary payload bytes, valid empty detection, invalid image/base64/unsupported options and operator limits, unchanged report parity, clean shutdown and the package file inventory. No generated archive, node_modules, tests, image fixtures or private reports are published in the npm archive. The source, runtime entry, schema and AGPL license are included.
Adapter source: integrations/mcp. Source engine: v0.11.0. The adapter is versioned separately from the engine. A local packed-consumer pass qualifies the tested archive and host; public registry installation and additional platforms retain separate checks. License: AGPL-3.0-or-later, including the adapter; see LICENSE.
SDK API sources: official TypeScript SDK, stdio server guidance, server metadata specification.