本站提供正體中文版。切換到正體中文本站提供简体中文版。切换到简体中文このサイトには日本語版があります。日本語で表示이 사이트는 한국어로도 제공됩니다.한국어로 보기Diese Website ist auch auf Deutsch verfügbar.Auf Deutsch ansehenEste sitio web también está disponible en español.Ver en españolQuesto sito è disponibile anche in italiano.Visualizza in italianoCe site est également disponible en français.Afficher en françaisEste site também está disponível em português.Ver em portuguêsDeze website is ook beschikbaar in het Nederlands.In het Nederlands bekijkenЭтот сайт также доступен на русском языке.Смотреть на русскомयह वेबसाइट हिन्दी में भी उपलब्ध है।हिन्दी में देखेंهذا الموقع متاح أيضًا باللغة العربية.عرض بالعربيةSitus ini juga tersedia dalam bahasa Indonesia.Lihat dalam bahasa IndonesiaBu site Türkçe olarak da mevcut.Türkçe görüntüleTa strona jest dostępna także po polsku.Wyświetl po polskuTrang web này cũng có phiên bản tiếng Việt.Xem bằng tiếng Việtاین وب‌سایت به فارسی هم در دسترس است.مشاهده به فارسی

Wrapping TIPO Patent Search into an MCP and Hooking It into AI

AI2026.06

Today I wrapped the publicly available patent information from TIPO (Taiwan’s Intellectual Property Office) into an MCP and hooked it into AI.

The settings screen of the tipo-opd extension installed in AI, showing it enabled, developed by Evan Tsai, and interfacing with TIPO’s patent open-information API

The Trouble Wasn’t the Features, but the Gap Between Documentation and Reality

Originally I just wanted to streamline the query workflow I use regularly, but once I actually got my hands dirty I found that the trouble often isn’t the functionality itself, but the gap that still exists between the official documentation and the results actually returned.

Along the way I also stepped into a few environment-configuration pitfalls. In particular, after packaging it into an .mcpb and installing it into AI, everything tested fine in the terminal, yet it kept showing “Server disconnected.” In the end I confirmed that the culprit was a difference in the environment path used when launching the MCP.

Conversational Queries Really Are a Lot More Convenient

At this point, I can already query a case’s basic information, related cases, and correspondence history directly within AI, and can also download PDFs of published documents.

The tool itself isn’t all that complex, but condensing a process that used to require repeatedly opening the website, entering criteria, and clicking through page by page into a single line of conversation really does make it quite a bit more convenient in practice.

Wait a minute—the more I play around with AI, the bigger the game seems to get. How did I unwittingly turn into an MCP developer?…