名刺をMCP化した話 — LINEに送った名刺を、ClaudeやChatGPTから自然言語で呼び出せるようにするまでThe author built a business-card manager called Meishibako that uses AI-OCR via…

A Japanese developer has published a technical write-up describing Meishibako, a business-card management service that combines AI-powered optical character recognition with the Model Context Protocol (MCP) so that saved contacts can be retrieved conversationally from assistants such as Claude and ChatGPT. The project matters because it illustrates a practical, low-friction way to turn a familiar analog artifact, the paper business card, into structured data that AI tools can actually query.

The workflow starts with a channel most people in Japan already use daily. A user photographs a business card and sends the image to Meishibako through LINE, the widely adopted messaging platform. According to the author, an AI-OCR step then reads the card within seconds, extracting fields such as name, company, department, job title, phone number, email address, and postal address, before organizing and storing the parsed result. This part of the system is essentially a modern take on established card-scanning apps, which have long promised to digitize contacts from a photo.

The author is candid that the OCR-to-storage pipeline is not the novel part. The distinctive element, as framed in the write-up, is adding MCP support so that the stored contacts become reachable from MCP-compatible AI tools. Once connected, a user can search and call up their own cards using natural language rather than navigating a database interface or a dedicated app screen. The piece is presented less as marketing and more as a build log, explaining the system's structure and the reasoning behind choosing MCP.

Understanding why that choice is significant requires some background on the protocol itself. MCP is an open standard, introduced by Anthropic in late 2024, that defines how AI assistants connect to external data sources and tools. In practice, a developer exposes their service through an MCP server, and any MCP-aware client, such as Claude or certain ChatGPT configurations, can discover and invoke that server's capabilities. The appeal is standardization: instead of building a separate, bespoke integration for each assistant, a single MCP server can, in principle, be reused across multiple compatible clients. That design is what allows Meishibako's contact data to surface inside more than one AI tool.

The stated motivation, according to the article's opening, is the observation that business cards are something you cannot easily hand to an AI. A stack of cards, or even a conventional contacts app, tends to sit outside the context an assistant can see. By exposing the card data through MCP, the author appears to be trying to close that gap, letting the assistant treat personal contact records as a queryable source during a conversation. The claimed payoff is that the setup turned out to be more useful in practice than expected, though that assessment reflects the developer's own experience rather than an independent benchmark.

For readers weighing the approach, a few caveats are worth keeping in mind. AI-OCR accuracy can vary with image quality, layout, fonts, and languages, so extracted fields may still need verification, a limitation common to all card-scanning tools. Handling business-card data also raises privacy and personal-information considerations, since names, phone numbers, and addresses belong to third parties; the source excerpt does not detail how storage, consent, or security are managed. Pricing, availability outside the described LINE-based flow, and the specific models used are likewise not spelled out in the material provided here.

The project also fits a broader industry pattern. Since MCP's release, a growing ecosystem of community and vendor-built servers has emerged to connect assistants to files, databases, and third-party services, and the specification has drawn interest from multiple AI vendors as a way to reduce integration overhead. Meishibako is a modest, personal example of that trend applied to a very concrete problem, and it demonstrates how consumer messaging apps like LINE can serve as an accessible front door for data capture while MCP handles the retrieval side. Whether this particular pattern scales into a mainstream contact-management workflow is not yet clear, but as a proof of concept it shows how existing OCR techniques and an emerging interoperability standard can be combined to make everyday information available to conversational AI.