Document Systems
A stateless MCP server for composing, validating, auditing, and rendering consequential documents from a small semantic model.
README
Document Systems
Experimental — not submitted to the OpenAI Plugins Directory. Document Systems is a public research prototype with verified engineering properties and unproven product utility. Do not treat its validation receipts or release gates as evidence that a document is correct, fit for use, or safer than an ordinary review.
Document Systems is a proposal-scoped Codex plugin prototype for composing, validating, gating, packaging, auditing, and rendering consequential documents from a small semantic model. It is published as a Banks Inc. service publication at documents.banksinc.us.
It packages four skills and a stateless MCP server. The server writes nothing, accepts nothing, and cannot make a document authoritative or operative.
Capabilities
- Compose immutable, content-addressed document Expressions.
- Validate structure, profile conformance, evidence references, status separation, caller-reviewed semantic invariants, and supplied lineage.
- Compile one-call release summaries or portable content-addressed packets.
- Verify a portable packet after caller-managed storage or transfer.
- Render traceable human-review, human-action, or machine projections.
- Audit authority, exceptions, evidence, readiness, and acceptance boundaries.
The v0.1 profile catalog contains:
normativeoperationalresearchhuman-agent
Each Expression selects exactly one primary profile and at most one secondary profile.
Install
Add this GitHub marketplace:
codex plugin marketplace add bankszach/document-systems
Install the plugin:
codex plugin add document-systems@document-systems-public
Start a new Codex session, then invoke a skill explicitly:
$compose-document-system Turn this operating requirement into a proposal Expression.
The plugin requires Node.js and uses no third-party runtime packages.
Tools
| Tool | Purpose |
|---|---|
list_document_profiles |
Return the four compiled profiles and their digests. |
compose_document_expression |
Normalize explicit semantic inputs into a deterministic proposal Expression. |
validate_document_expression |
Return a scoped structural receipt for an Expression and supplied lineage. |
render_document_view |
Render a bounded, traceable projection without writing an artifact. |
compile_document_packet |
Compose, validate, semantically gate, and render in one call; return a summary or portable packet. |
verify_document_packet |
Recompute a portable packet's identity, receipt, view, and release gate. |
All tools advertise readOnlyHint: true, openWorldHint: false, and
destructiveHint: false, and each declares an output schema.
The public stateless Streamable HTTP endpoint is:
https://documents.banksinc.us/mcp
Safety boundary
An Expression is always PROPOSED. Validation can establish structural
properties and conformance to caller-reviewed semantic assertions, but it does
not independently establish:
- source truth or source authority;
- behavioral success or intended-use fitness;
- human acceptance or operative status;
- repository head or complete lineage;
- persistence of returned objects.
Every returned object declares persistence_status: NOT_PERSISTED. A FULL
release packet is portable and content-addressed so a caller can store and
later verify it; the service itself does not store or resolve packets.
Validate
npm test
The test suite exercises deterministic composition, content tampering, profile drift, supplied lineage, evidence laundering, lifecycle contamination, profile overcomposition, fail-closed action rendering, HTML escaping, element-level traceability, semantic drift, typed allocation completeness, portable-packet verification, and the no-write boundary.
Run the external OSHA, CISA, CDC, and NIST benchmark:
npm run benchmark:real-world
See Real-World Benchmark v0.2 for the measured self-authored structural and semantic-defect detection, payload reduction, and why those results do not establish marginal product value.
See Experimental Evaluation v0.3 for the World Control Plane failure that caused the production claim to be withdrawn and the evidence required before reconsidering public submission.
Distribution status
This repository is a public GitHub marketplace for Codex and contains a public experimental HTTPS MCP transport, policy pages, test cases, and dormant submission materials. The plugin has not been submitted to OpenAI for review. Banks Inc. will not submit or publish it in the universal Plugins Directory unless a blinded baseline-versus-candidate evaluation demonstrates repeatable marginal value.
License
MIT
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。