CiteWire
Enables agents to query news and research metadata through MCP tools, either by wrapping a compatible news platform or enabling public provider adapters, while preserving attribution and publisher credit.
README
CiteWire
Attribution-first MCP infrastructure for news and research discovery.
CiteWire gives agents structured, read-only access to news and research metadata while preserving credit and traffic for the original publisher. It can expose a compatible news platform as typed MCP tools, query optional public provider APIs, or do both in one server.
The Community edition is useful on its own. It includes the MCP core, stdio and HTTP transports, the generic platform adapter, and every provider adapter in this repository under the MIT license. Commercial product direction is additive and focused on managed operations, personalization, workflows, collaboration, retained history, and support. See Product tiers.
What CiteWire is for
- News platform operators can expose a compatible read API through
news.list,news.get,news.topics, andnews.about. - Researchers and agent builders can enable only the public news and research providers they want to query.
- Self-hosters can run the same server over stdio, stateless HTTP, or inside a Node serverless function.
- Provider contributors can add a focused adapter without introducing a runtime dependency.
CiteWire is not a full-text ingestion service, crawler, publisher, editorial system, or rights-clearance service. It does not store or republish article bodies. Provider reachability does not replace review of the provider's current terms for your use case.
Quick start
Requirements: Node.js 18 or newer. Community is an ESM-only Node package; it does not provide a CommonJS entry point.
Create citewire.config.json with one provider enabled:
{
"providers": {
"openalex": { "enabled": true }
}
}
Run the server over stdio:
npx -y citewire --config citewire.config.json
stdio is the default transport and is the usual choice when an MCP client launches citewire as a local process.
MCP client configuration
Use an absolute config path because clients do not always launch from the project directory:
{
"mcpServers": {
"citewire": {
"command": "npx",
"args": [
"-y",
"citewire",
"--config",
"/absolute/path/to/citewire.config.json"
]
}
}
}
HTTP transport
Run the stateless HTTP transport on port 8722:
npx -y citewire --config citewire.config.json --http 8722
The local endpoint accepts JSON-RPC over POST /. It does not offer sessions
or an event stream.
Two ways to use it
1. Wrap a compatible news platform
Declare a platform with a display name, site URL, and read API base:
{
"platform": {
"name": "Example Industry News",
"siteUrl": "https://news.example",
"apiBase": "https://news.example/api/v1/news"
}
}
That configuration exposes four tools:
| Tool | Purpose |
|---|---|
news.list |
List items newest first, with optional topic, industry, text, date-window, and pagination filters. |
news.get |
Get one item by slug, including attribution and related coverage returned by the platform. |
news.topics |
Read the platform's active topic taxonomy. |
news.about |
Read static platform facts and the attribution policy. |
The required HTTP surface is documented in the platform read-API contract.
2. Enable public provider tools
Provider adapters are included in Community and are disabled by default. A
provider becomes active only when its config entry sets enabled to true.
{
"providers": {
"gdelt-doc": { "enabled": true },
"arxiv": { "enabled": true },
"crossref": {
"enabled": true,
"mailto": "operator@example.com"
}
}
}
Crossref requires the deployer's own contact email so requests can identify the
operator to its polite pool. citewire sends that address only to Crossref in
the request mailto parameter and User-Agent. Do not copy the example
address.
The current adapters are:
| Provider | Tools | Focus |
|---|---|---|
| GDELT DOC 2.0 | gdelt.search |
Worldwide news article metadata |
| GDELT Context 2.0 | gdelt.context |
Snippet-level context around a term |
| arXiv | arxiv.search |
Preprint metadata |
| OpenAlex | openalex.search |
Scholarly works, authors, and venues |
| Crossref | crossref.search |
DOI registration metadata |
| Semantic Scholar | semanticscholar.search |
Papers and author graph metadata |
| Europe PMC | europepmc.search |
Life-sciences literature metadata |
| dblp | dblp.search |
Computer-science bibliography records |
| Hacker News | hackernews.top, hackernews.item |
Stories and items from the official API |
| DEV | devto.search |
Published DEV article metadata |
Read Providers before enabling one. That document records the endpoint, documentation, free-access basis, and known courtesy limits for each adapter. The provider's own documentation remains the source of truth.
Configuration behavior
platformis optional.providersis optional.- A provider omitted from the config remains disabled.
- An empty config is valid and exposes no tools.
- Invalid config fails at startup with the field named in the error.
tools/listis the authoritative tool list for a running deployment because enabled tools depend on that deployment's config.
A deployment can combine a platform and provider tools:
{
"platform": {
"name": "Example Industry News",
"siteUrl": "https://news.example",
"apiBase": "https://news.example/api/v1/news"
},
"providers": {
"openalex": { "enabled": true }
}
}
Optional Community policy foundation
CiteWire can expose a local, read-only foundation for source policy, rights evaluation, and shadow inclusion scoring:
{
"community": {
"enabled": true,
"classifierMode": "shadow",
"thresholds": {
"adjacent_min": 0.5,
"standard_min": 0.6
}
}
}
Every bundled source remains disabled by default. This setting performs no
network calls, loads no credentials, and cannot publish. Credential-based
rights checks accept opaque secret-manager references in memory but never
return those references. The classifier is forced to shadow mode and always
reports publishable: false until a separately reviewed editorial system and
evaluation corpus exist.
Read Source registry and rights for the runtime validation, fail-closed policy, canonical MCP tools and resources, and current limitations.
Design boundaries
- MIT and zero dependency. The package uses Node built-ins and has no runtime or development dependencies.
- Node ESM. Community supports Node.js 18 or newer through ESM. It does not provide CommonJS compatibility or a CommonJS build.
- Read only. The included tools query platform and provider read surfaces.
- Stateless. The Community server has no accounts, sessions, saved searches, or retained history.
- Attribution first. Results preserve source metadata and links supplied by the upstream platform or provider.
- Config driven. A deployer decides which tools are present. Nothing turns itself on.
- Storage free. The core does not retain provider responses or source content.
These boundaries are maintained through the project governance policy.
Ecosystem and product documentation
- Product tiers defines the permanent Community value and the additive commercial boundary.
- Distribution contains the exact registry and catalog
checklist. Unperformed submissions are marked
PENDING. - Governance explains decisions, provider stewardship, compatibility, and licensing.
- Providers documents each upstream provider.
- Platform contract defines the compatible news read API.
- Contributing covers setup, tests, provider additions, and pull request expectations.
- Releasing defines package and registry release controls.
- Support, Security, and the Code of Conduct define community operating channels.
Project origin
citewire was shaped through its first deployment for Karaya Group Industry News. The public platform contract is generic and can wrap any compatible read API.
Contributing
Run the test suite with npm test. No install step is required. See
CONTRIBUTING.md before proposing a provider or a change to a
public contract.
License
Community is MIT licensed. See LICENSE.
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。