lex-provenance-mcp
A read-only MCP connector for searching, fetching, and citing provenance-tracked legal corpora with verifiable content hashes.
README
LexProvenance - lex-provenance-mcp
Verifiable, provenance-first legal corpora for AI systems. A read-only Model Context Protocol connector that lets AI agents search, fetch, and cite neutral, provenance-tracked bodies of law - with a tamper-evident content hash on every citation.
Status: v0.1 - public scaffold. This is a deliberately thin, credible interface layer, not a finished product. See the roadmap and what this is not.
Why this exists
A dataset says "here are some documents." This connector says "here is how an AI agent can search, fetch, and cite these documents safely, with each answer traceable to an official source and verifiable against a content hash."
The connector is corpus-agnostic. It speaks to any provenance-first legal corpus that exposes a standard manifest - today three public "global-commons" corpora, tomorrow any body of law or regulation. The connector is the interface; the corpora are the asset.
What it connects (today)
| Corpus | Domain | Canonical home |
|---|---|---|
space_law |
International & national space law | github · Hugging Face dataset |
deep_seabed_mining |
Deep seabed mining law (UNCLOS Part XI, ISA, US regime) | github · Hugging Face dataset |
bbnj_high_seas |
BBNJ / High Seas Treaty & framework | github · Hugging Face dataset |
This repo ships only sample manifests (a few records each) that point to those canonical sources. It does not contain the full corpora.
Tools (read-only)
| Tool | What it does | Tier |
|---|---|---|
list_corpora |
List available corpora + coverage | Free |
search_corpus |
Keyword search within one corpus | Free |
fetch_document |
Return one document's metadata + provenance | Free |
fetch_provision |
Return one provision (article/section) | Free |
get_citation |
Return a verifiable citation (source, date, SHA-256 hash, version) | Free |
compare_provisions |
Relate/compare provisions across corpora | Stub -> premium |
Every tool is read-only. There are no write, edit, ingest, delete, or shell tools - by
design, and enforced by a test (tests/test_read_only.py). See docs/security.md.
Verifiable citations (the differentiator)
get_citation returns the content SHA-256 hash and the dated version of the exact text
it cites. An agent's citation therefore points not just at "Article 14" but at a specific,
reproducible, tamper-checkable state of Article 14. If the stored text changes by one
character, the hash changes - so a citation can be re-verified later. See
docs/citation-model.md.
Quickstart
pip install -e ".[dev]" # pure-Python tool logic + tests, no MCP SDK needed
python -m pytest -q # all green
# To actually serve over MCP (stdio):
pip install -e ".[mcp]"
lex-provenance-mcp # or: python -m lex_provenance_mcp.server
Try the tool logic directly (no SDK required):
from lex_provenance_mcp import tools
print(tools.list_corpora())
print(tools.search_corpus("liability", corpus_id="space_law"))
print(tools.get_citation("space_law", document_id="outer-space-treaty", provision="Article VI"))
Commercial boundary
This connector is the free razor. The enriched, connected, and hosted layers are a separate
paid offering - see docs/commercial.md. In short:
- Free / open: the connector, schemas, sample manifests, docs, in-corpus search/fetch/cite.
- Not in this repo (premium): full enriched corpora, cross-corpus reasoning, embeddings, monitoring / treaty-status intelligence, hosted search, private-corpus onboarding.
What this is not
- Not the corpora (those live in their own repositories and datasets).
- Not the corpus-construction methodology (that is a separate, private toolkit - this repo reveals what you fetch, never how the record is built).
- Not legal advice. See
docs/legal-disclaimer.md.
Roadmap
- v0.1 (this): read-only connector scaffold, schemas, sample manifests, in-corpus tools, verifiable citations, adapter placeholders. No hosted service, no full corpus, no premium layer.
- v0.2: local search over selected public corpora; adapter examples that actually run.
- v0.3: richer citation model + corpus manifests consumed directly from canonical repos.
- Later (premium, not here): hosted MCP endpoint, cross-corpus intelligence, monitoring, institutional access.
Renaming
The public identity lives in one file - mcp_server/identity.py.
Change BRAND_NAME, REPO_SLUG, TOOL_NAMESPACE, PACKAGE_NAME there (and rename the folder)
to switch to any alternative name; nothing else hard-codes it.
Licence
Code and schemas: Apache-2.0 (LICENSE). Docs: CC BY 4.0 unless stated.
Source legal texts are official public documents and are never relicensed. See NOTICE.
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。