hazmat-cfr-mcp
Enables AI agents to look up and validate hazardous materials shipping descriptions using public 49 CFR citations, providing structured JSON with proper shipping names, hazard classes, labels, and regulatory references.
README
hazmat-cfr-mcp
The complete 49 CFR 172.101 Hazardous Materials Table, as an MCP tool for AI agents — every field cited to a pinned eCFR revision.
Give Claude, Cursor, Codex, and other MCP clients a DOT hazmat lookup, classification, and validation tool backed by the entire Hazardous Materials Table — not a sample.
Ask an agent:
Can I ship 60 L of acetone on a cargo-only aircraft? What are its label, packing, and special-provision requirements?
It returns the proper shipping name, hazard class, packing group, label codes, packaging references, aircraft limits, vessel stowage, decoded special provisions, and CFR citations — in structured JSON.
Use it two ways:
hazmat-cfr-mcp— MCP stdio server for agentshazmat-cfr— local CLI for humans, tests, and scripts
No API keys. No customer data. No carrier contracts. No NMFC. Public CFR only.
What's in the box
- ~2,700 Hazardous Materials Table entries (49 CFR 172.101) — every UN/NA entry plus Forbidden materials, ingested directly from the official eCFR.
- Special-provision decoding — all ~420 Column 7 codes (49 CFR 172.102: numeric, A, B, IB, IP, N, R, T, TP, W) expanded into their regulatory text.
- Legend decoding — Column 1 symbols (
+ A D G I W), Column 10A vessel stowage categories, and Columns 8A–8C packaging references resolved to part-173 sections. - Pinned to a dated eCFR snapshot so every answer is reproducible and citable to an exact revision.
- A curated synonym/CAS layer so "IPA", "muriatic acid", or "bleach" resolve — kept clearly separate from regulatory data.
10-second try
npx -y hazmat-cfr-mcp --help
npx -y hazmat-cfr lookup UN1088 # not in any "sample" — the whole table is here
npx -y hazmat-cfr decode-sp IB2 T8 A3
Why this exists
Foundation models are weak at regulated shipping details. They may know what acetone is, but they should not guess whether a shipping description is complete, what IB2 means, or which CFR field backs a label requirement.
This server gives agents an explicit, cited, complete hazmat tool surface so they answer with the regulation instead of a hallucination.
Accuracy & verification
This is a compliance-adjacent tool, so correctness is enforced by the build, not by hope:
- The dataset is generated from the eCFR versioner API and committed to the repo (see docs/data-pipeline.md). It is never hand-edited.
- The ingest fails if any special-provision code referenced by the table does not resolve in 172.102 (documented source errata, e.g.
IP16, are resolved to an explicit "referenced but not defined" note — never fabricated). test/data-integrity.test.tsre-checks entry counts, structural validity, referential integrity, and ~22 hand-verified spot-checks on every run.- The parsers are unit-tested against fixtures in
test/fixtures/.
⚠️ Disclaimer. This is an informational aid that surfaces public CFR text. It is not legal advice, not a compliance certification, and not a substitute for a trained hazmat employee. Always verify any shipping decision against the current eCFR.
Install and run
npx hazmat-cfr-mcp # MCP stdio server
npx hazmat-cfr lookup UN1090 # CLI
From a clone:
git clone https://github.com/andretaki/hazmat-cfr-mcp.git
cd hazmat-cfr-mcp
npm install
npm run build
npm run demo
CLI:
npx hazmat-cfr lookup UN1090
npx hazmat-cfr validate "UN1090, Acetone, 3, PG II"
npx hazmat-cfr segregation 3 5.1 8
npx hazmat-cfr lq 3 II 1000 12 true
npx hazmat-cfr decode-sp IB2 T8 A3
npx hazmat-cfr decode-pkg nonBulk 202
MCP client config
{
"mcpServers": {
"hazmat-cfr": {
"command": "npx",
"args": ["hazmat-cfr-mcp"]
}
}
}
For a local checkout, point command at node and args at dist/server.js.
Tools
| Tool | Purpose |
|---|---|
lookup_hazmat_entry |
Look up by UN/NA number or proper shipping name; returns matching entries with decoded symbols and special provisions. |
classify_shipping_description |
Parse free text into structured fields, then validate it. |
validate_basic_hazmat_description |
Validate structured fields against the full 172.101 table. |
get_label_requirements |
Return hazard label codes and caveats. |
check_basic_segregation |
Conservative pairwise segregation findings (49 CFR 177.848 subset). |
check_limited_quantity_eligibility |
Conservative Class 3/Class 8 LQ-candidate screen. |
decode_special_provision |
Expand Column 7 codes (e.g. IB2, T8, A3) into 49 CFR 172.102 text. |
decode_packaging_reference |
Resolve a Column 8A/8B/8C value to its part-173 section. |
explain_cfr_source |
Explain which public CFR source backs a field or rule. |
Example call:
{ "query": "UN1090" }
Library use
import { defaultCatalog, decodeSpecialProvisions, ECFR_SNAPSHOT_DATE } from "hazmat-cfr-mcp";
const acetone = defaultCatalog.lookup("UN1090");
const provisions = decodeSpecialProvisions(acetone.entries[0].specialProvisions);
console.log(`Backed by eCFR snapshot ${ECFR_SNAPSHOT_DATE}`);
Refreshing the data
# bump ECFR_SNAPSHOT_DATE in src/data/snapshot.ts, then:
npm run ingest -- --fresh
npm test
See docs/data-pipeline.md for the full pipeline.
Non-goals
Not legal advice; not a replacement for trained hazmat employees; not a full compliance certification engine; not NMFC classification; not carrier contract/tariff analysis. No private customer data, contract rates, or internal business records belong in this repo.
Safety
npm test
npm run typecheck
npm run scan:secrets
Maintainer
Andre Taki — Alliance Chemical — andre@alliancechemical.com
License
MIT — the underlying 49 CFR text is a public-domain work of the U.S. government.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。