doc-tools-mcp
An MCP server for managing pandoc/CSL-YAML bibliographies and verifying citations in documents.
README
Doc Tools MCP Server
⚠️ Work in Progress — APIs may change without notice.
A Model Context Protocol (MCP) server for managing pandoc/CSL-YAML bibliographies and verifying citations. Lets your LLM read, search, add, update, delete, validate bibliography entries, and systematically verify that citations support the claims made in your text.
Quick Start
npx @metaneutrons/doc-tools-mcp
Add to your MCP client config (e.g., claude_desktop_config.json):
{
"mcpServers": {
"doc-tools": {
"command": "npx",
"args": ["-y", "@metaneutrons/doc-tools-mcp"],
"env": {
"DT_BIB_YAML": "/path/to/references.yaml",
"DT_BIB_CSL": "/path/to/style.csl",
"DT_CT_REGISTRY": "/path/to/ctverify.json"
}
}
}
}
Environment Variables
| Variable | Description |
|---|---|
DT_BIB_YAML |
Default path to CSL-YAML bibliography file. Makes file param optional on all bib:* tools. |
DT_BIB_CSL |
Default path to CSL style file (.csl). Enables style-aware field validation on bib:validate, bib:add, bib:update. |
DT_CT_REGISTRY |
Default path to citation verification registry JSON. Makes registry_path optional on all ctverify:* tools. |
All env vars are ignored if the referenced file does not exist. Tool descriptions adapt dynamically to show configured defaults.
Bibliography Tools (bib:*)
Read
| Tool | Description |
|---|---|
bib:get |
Retrieve a single entry by ID (full YAML block) |
bib:search |
Full-text search across all fields: author, title, type, year, editor, container-title |
bib:list |
List all entries of a given type (e.g., legal_case, chapter, article-journal) |
bib:exists |
Check if an ID exists (fast boolean check before citing) |
bib:stats |
Entry count total and breakdown by CSL type |
Write
| Tool | Description |
|---|---|
bib:add |
Add a new entry with duplicate ID check, required field validation, and YAML confirmation output |
bib:update |
Patch individual fields of an existing entry (other fields remain untouched) |
bib:delete |
Remove an entry by ID |
Validation
| Tool | Description |
|---|---|
bib:validate |
Check entire file: YAML syntax, required fields per CSL type, duplicate IDs, missing issued dates |
All write operations automatically create a .bak backup before modifying the file.
CSL Style Validation
When a CSL style file is configured (via DT_BIB_CSL or the style parameter), validation uses the variables actually referenced in the style instead of hardcoded required fields. Entry types not handled by the style produce a warning.
Without a style file, these hardcoded defaults apply:
| Type | Required Fields |
|---|---|
legal_case |
title, authority, number, issued |
book |
title, issued |
article-journal |
title, container-title, issued |
chapter |
title, container-title, issued |
legislation |
title, issued |
thesis |
title, issued |
Citation Verification Tools (ctverify:*)
Systematic verification that cited sources actually support the claims made in the text.
Workflow: extract → set claims → verify each claim against the source → update status.
| Tool | Description |
|---|---|
ctverify:extract |
Extract citations from Pandoc inline footnotes (^[...]). Captures ~200 chars of context before each footnote. Merges into registry preserving existing claims/statuses. Supports dry_run to preview changes after text edits. |
ctverify:update |
Batch update citation entries: set claim, status, note per entry via entries array. Creates new entries when cite is provided. |
ctverify:bulk-update |
Batch status update by ID list or filter_status. |
ctverify:status |
Show verification progress with counts. Filter by filter_status, filter_claim (use "" for missing claims), or file (basename/suffix match). |
Re-sync after edits
Re-running ctverify:extract after text changes is safe:
- Exact match — citations with unchanged text keep their claims and status
- Proximity match — if cite text changed but the position is within ±5 lines, claims are recovered from the nearest previous entry
dry_run: true— preview new, removed, and proximity-matched entries before committing
Architecture
src/
├── index.ts # MCP server, dynamic provider loading, stdio transport
├── shared/
│ ├── types.ts # Provider, ToolDefinition, ToolResult interfaces
│ ├── logger.ts # Structured logging (pino → stderr)
│ └── errors.ts # Typed error hierarchy
└── providers/
├── bib/ # Bibliography provider
│ ├── index.ts # BibProvider (9 tool handlers, env var resolution)
│ ├── store.ts # YAML read/write with .bak backup
│ ├── schema.ts # Zod schemas + per-type field validation
│ ├── csl-style.ts # CSL style parser (variable extraction per type)
│ └── tools/ # Dynamic tool definitions (read + write)
└── ctverify/ # Citation verification provider
├── index.ts # CtverifyProvider (4 tool handlers)
├── extract.ts # Pandoc inline footnote parser
└── types.ts # CitationEntry interface
The provider system is extensible — add a new directory under src/providers/ with a createProvider() export and it will be auto-discovered at startup.
Development
npm install # Install dependencies
npm run build # Compile TypeScript
npm test # Run tests
npm run test:watch # Watch mode
npm run test:coverage # Coverage report
npm run lint # ESLint
Commit Convention
This repo uses Conventional Commits enforced via Husky + commitlint.
Types: feat, fix, docs, style, refactor, perf, test, chore, ci, build, revert
Scopes: bib, ctverify, core, deps, config
License
GPL-3.0 — See LICENSE for details.
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。