Context7
Up-to-date docs for AI agents
作者提供的说明(英文原文)
Look up, search, and track changes to the US Code of Federal Regulations (eCFR).
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": {
"cfr": {
"command": "uvx",
"args": [
"cfr-mcp"
]
}
}
}在终端中运行这条命令
claude mcp add cfr -- uvx cfr-mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"cfr": {
"command": "uvx",
"args": [
"cfr-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": {
"cfr": {
"type": "stdio",
"command": "uvx",
"args": [
"cfr-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7日
注册名称已通过 GitHub 账户 mccallar 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(0.2.2)于2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 Python (uv)。
同一分类中有:Context7, Codebase Memory, Exa
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
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
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。
An MCP server that gives AI assistants access to the US Code of Federal Regulations.
Ask "what does 21 CFR 101.9 require?" or "has 40 CFR 261 changed since 2023?" and get the actual regulation text, with citations, instead of a plausible-sounding guess.
Unofficial community project. Not affiliated with or endorsed by the Office of the Federal Register, NARA, or GPO, and uses no government seals or logos.
Maintained by one person as a personal project and provided as-is (MIT). Issues and pull requests are welcome but may not get a quick response. To report a security issue, see SECURITY.md.
Requires Python 3.11+. No API key — the eCFR and Federal Register APIs are open.
Claude Code:
claude mcp add cfr -- uvx cfr-mcp
Any other MCP client:
{
"mcpServers": { "cfr": { "command": "uvx", "args": ["cfr-mcp"] } } }
| Tool | What it does |
|---|---|
lookup_citation | Text of a citation — 21 CFR 101.9, 40 CFR 261.4(b)(1), 40 CFR Part 261 Subpart C |
search_regulations | Full-text search; returns citations and snippets, never bodies |
where_does_term_appear | Which titles contain a term, with hit counts. Fetches no text |
browse_structure | The hierarchy of a title or part, no text |
what_changed | Amendment history for a citation, linking each change to the Federal Register rule that caused it, plus published corrections |
compare_versions | Diff of a citation's text between two dates — see exactly what an amendment changed |
list_agencies | Maps agency names to the CFR titles they administer |
Point-in-time works throughout: pass date as YYYY-MM-DD to read the CFR as it stood.
Context budget is the whole game. The eCFR full endpoint returns an entire downloadable XML document for a title-level request. Every tool here caps output and degrades to an outline rather than dumping text into the model's context. Title-level XML requests are refused outright.
Dates must be resolved, not assumed. Versioner routes 404 on dates that aren't valid issue dates for a title, so the client resolves through titles.json first rather than passing today's date blindly.
Caching is courtesy. The eCFR publishes no rate limit and has no key to identify callers politely, so the client caches to disk (historical dates forever, since point-in-time content is immutable) and self-limits concurrency. The cache lives in ~/.cache/cfr-mcp (respects XDG_CACHE_HOME); set CFR_MCP_CACHE_DIR to relocate it.
Regulation text is free to reproduce. 1 CFR 2.6 states that any person may reproduce or republish material appearing in the Federal Register, with no restrictions on what is reproduced, who reproduces it, or where. Federal government works are also outside copyright under 17 U.S.C. §105.
Incorporation by reference. Some CFR sections incorporate private standards (ASTM, NFPA, ASHRAE) whose copyright status after incorporation remains unsettled. The eCFR does not contain the text of those standards and neither does this server — it returns the citation only. Obtain standards from the issuing organization or the Office of the Federal Register reading room.
Status of the text. The eCFR is authoritative but unofficial. Anyone relying on it for legal research should verify against the current official CFR, the daily Federal Register, and the List of CFR Sections Affected (LSA).
Not legal advice. This is a retrieval tool. It returns the text of regulations; it does not tell you whether you are compliant with them.
uv sync --extra dev
uv run pytest
MIT licensed.