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)
MCP server for compare-cli: contract version diff/redline and a drift gate as agent 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。
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": {
"compare-cli": {
"command": "npx",
"args": [
"-y",
"compare-cli-mcp"
]
}
}
}在终端中运行这条命令
claude mcp add compare-cli -- npx -y compare-cli-mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"compare-cli": {
"command": "npx",
"args": [
"-y",
"compare-cli-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": {
"compare-cli": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"compare-cli-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 DrBaher 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(0.1.4)于5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: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。
MCP (Model Context Protocol) server wrapping compare-cli for agent pipelines. Three tools, stdio transport, JSON-first responses with optional human report.
Design contract:
../docs/mcp.md. This package ships the implementation; behavior and schemas track the design version.
Part of the contract-ops suite.
compare-cli-mcp has compare-cli as a peer dependency, so install both:
npm install -g compare-cli@^0.3.0 compare-cli-mcp@^0.1.1
Then run:
compare-mcp # spawns the MCP server on stdio
Add to your MCP-client config (e.g. ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": { "compare-cli": { "command": "compare-mcp" } } }
{
"mcpServers": { "compare-cli": { "command": "compare-mcp", "env": { "COMPARE_MCP_BASE_DIR": "/path/to/contract/documents" } } } }
Every path argument the agent supplies is then resolved against COMPARE_MCP_BASE_DIR and rejected if it escapes the directory (symlinks collapsed). Without the env var the server reads any file the process can read — see the security note in ../docs/mcp.md §3.2.
| Tool | Purpose |
|---|---|
compare_files | Two-document comparison. Each side accepts path, content_base64, or content_text. |
compare_with_negotiation | Base is read from nda-review-cli's negotiation.json. Returns base.negotiation_resolution so the agent can audit which signal produced the agreed text. |
compare_demo | Zero-arg synthetic comparison against bundled fixtures. Deterministically returns exit_class: "substantive". Use to confirm the server is wired up. |
Tool descriptions, input schemas, and output schemas are advertised on tools/list. The schemas are stable across v1.x of this package; any change is fenced by the snapshot golden in tests/fixtures/tools-list.json.
Every successful tool call returns:
compare --json shape (top-level ok, exit_class, exit_code, base, candidate, summary, differences, warnings). Stable across compare-cli v1.x per its AGENTS.md. This is the contract the agent reads.structuredContent.include_human_report: true) — the human-readable report.compare_with_negotiation adds structuredContent.base.negotiation_resolution ∈ {"status", "per_round_agreed", "clause_status"} indicating which of the three resolution tiers produced the agreed base text.
Substantive drift is a successful tool call. The agent asked what's different and the server answered. Routing on exit_class is the contract — error envelopes (isError: true) are reserved for I/O / argument failures plus one explicit precondition exception:
| Code | Trigger |
|---|---|
INPUT_NOT_FOUND | Path not found / unreadable. |
INPUT_MALFORMED | Malformed .docx / .pdf. |
PDF_NO_TEXT_LAYER | Scanned PDF; pdfjs extracted zero characters. |
INVALID_ARGS | Schema-validation failure (missing args, oneOf violation, etc.). |
PATH_OUTSIDE_BASE_DIR | COMPARE_MCP_BASE_DIR is set and the path escapes it. |
NO_AGREED_ROUND | compare_with_negotiation precondition: no agreed round found. The CLI returns exit 2 here; the MCP server promotes it to an error because no comparison ran. |
NOT_SIGNED_OFF | require_signoffs: true and signoffs.a / signoffs.b is missing. |
INTERNAL_ERROR | Unexpected; surfaces underlying message so the agent isn't blind. |
Codes are stable across v1.x of this package.
# From the repo root:
npm install # workspaces resolve compare-cli + compare-cli-mcp cd mcp && npm test # 22 tests, ~100 ms
The conformance + snapshot suite is in tests/test_protocol.mjs. The tools/list golden lives at tests/fixtures/tools-list.json. To regenerate the golden after an intentional schema change (also bump the minor version):
node -e 'import("./compare-cli-mcp.mjs").then(m => process.stdout.write(JSON.stringify(JSON.parse(JSON.stringify(m.TOOLS)), null, 2)))' > tests/fixtures/tools-list.json
MIT. See ../LICENSE.