The Central News Agency launches an MCP to connect 5 million pieces of verified news to AI
Taiwan’s media industry’s first case! Taiwan’s public media organization (Central News Agency) has recently launched the “Central News Agency MCP” service and opened subscriptions (note), becoming Taiwan’s first news media to offer MCP for AI agents to access.
Users subscribed to the Central News Agency MCP do not need to change their original AI usage habits. By simply subscribing and attaching the tool to their AI, they can bring in (Central News Agency)’s accumulated news database of over 30 years and nearly 5 million items, enabling AI applications such as Claude and ChatGPT to instantly retrieve information when answering questions—and to require explicit source attribution.
Note: MCP is an open standard agreement introduced by Anthropic in November 2024. In Chinese, it is usually translated as the model context protocol. Its purpose is to let AI models access external data and tools in a safe, standardized way. Some also liken it to an AI USB: as long as the user’s preferred AI application provides a plug-in or connector feature, different MCPs provided by different companies can be installed to further connect the data needed.
What is CNA MCP? What functions does it offer?
CNA MCP integrates more than 30 years of nearly 5 million press releases, 3.5 million news photos across a century, a world yearbook database from 2000 to the present, and publicly available data from about 150 government agencies across Taiwan.
During AI Q&A, the AI will introduce CNA’s accumulated years of Traditional Chinese terminology conventions, and it can also call professional news organizations’ fact-checking tools to verify suspicious information.
Image source: CNA MCP official website What is CNA MCP? What functions does it offer?
Wang Zhijian, distinguished professor in the Graduate Institute of Information Management at National Taipei University, pointed out that MCP tools not only help AI identify correct information, but also help overcome language limitations so that users in other countries can access information. Xie Jilong, director of the Graduate Institute of Journalism at NTU, said that CNA MCP provides trustworthy and clear sources of information, making it a forward-looking development.
In terms of cost, CNA MCP’s monthly subscription price is NT$200. Current subscribers can enjoy an early-bird discount plan, and query limits are increased with the bonus.
(CNA) will also roll out an AskCNA service in the future, allowing users to ask questions directly on the official website. AI will answer in real time and include CNA citation links and the original text that can be traced back.
Where CNA MCP’s value lies: in archiving “old” news and giving it new worth
Jiaci He, a developer, shared on Facebook that a proof-of-concept version of CNA MCP was completed months ago. Over the past month, the team has continued to collaborate to integrate payment flows, strengthen cybersecurity, and rebuild back-end logic and the database design.
In general, MCP available on the market can be broadly divided into two types: tool-based and data-based. Tool-based MCP allows AI to use instructions to call existing databases to complete tasks for users. Data-based MCP has only gradually become popular in recent months; it builds data in a specific domain into an MCP. Its biggest advantage is that ordinary web searches cannot reach these data.
In the CNA database, news that is more current can mostly be found through online searches; the real value of MCP lies instead in long-established assets such as an archive of past news that has not yet been posted online, century-old photo collections, and world yearbooks.
In addition to the above, CNA MCP also has the following advantages and features:
Conventions for Traditional Chinese terminology: During AI Q&A, (CNA) accumulated years of Traditional Chinese language conventions are introduced so that AI responses are closer to Taiwan’s familiar translated names and people’s names, avoiding the misuse of terms translated from Chinese for use in Mainland contexts.
Fact-checking tool:You can call professional news organizations’ fact-checking processes to help verify suspicious information.
Keyword search mechanism: Based on (CNA)’s existing Elastic Search system, it uses keyword search and supplements semantic judgment with a Reranker to improve retrieval accuracy.Visual presentation of MCP Apps:Let AI not only respond with text, but also present information in forms such as charts and small websites—for example, tools like quick primers and timelines.
Skill function: Package (CNA)’s fact-checking process into an MCP call tool, teaching AI how to properly use CNA data.
What is the significance of CNA’s MCP for the media in the AI era?
The idea behind CNA MCP comes from a news MCP topic shared by Florent Daudens to the development lead Song Ka-po last September, when the “zero-click” phenomenon was just beginning to attract discussion.
After the development team, it received funding from Google Taiwan News Digital Co-Prosperity Fund (nDX Taiwan News Digital Innovation Program) to complete Taiwan’s first news MCP, and it may also be, after Reuters, the second news MCP globally released by an official news organization. For now, it will proceed with a trial program for the enterprise model. It is expected that formal charging will begin next year, and Korea’s Yonhap News Agency also plans to launch a similar service by the end of the year.
Song Ka-po pointed out that the news industry has expanded from serving businesses and consumers to serving AI agents; as the intermediary role shifts from humans to machines, how to maintain public trust in news has become an important issue: “In my view, CNA’s MCP is a century-old organization that, at this stage, we can present to people in Taiwan and even international friends—a window for understanding Taiwan and deepening context.”
He hopes that CNA MCP can help users regularly organize trend information, clarify changes in stances on specific topics, cross language barriers to obtain CNA-related reports, and quickly verify suspicious posts when users encounter them on social media.
Further reading:
CNA sues NTU student—what happened next! The two sides have reached a settlement after the student created a Traditional Chinese dataset for AI that allegedly involved infringement
“A first for Taiwanese media! CNA launches an MCP subscription service, bringing 5 million trustworthy news items to AI agents”—the article was first published on “Encrypted City”