Yiren Lu

Yiren Lu is a writer and software engineer whose AI work includes turning videos into documentation and, by her own account, authoring MCPKit, a blueprint for authenticated Model Context Protocol servers. Her practical arguments for GPT emphasize extracting structured information and connecting models to external data and systems.[2][1][5][3]

In a November 6, 2025 post, Lu described the MCPKit solution blueprint as “authored by yours truly.” MCPKit provides a sample repository for teams connecting data to ChatGPT or other MCP clients. Lu explained that organizations could fork the repository, change its tool definitions and contents, and integrate their own data and systems.[5][6]

Lu argued that access to external data and systems would bring the next major increase in ChatGPT's usefulness. She emphasized authentication as a way to restrict tools to users with the appropriate permissions—particularly relevant to the ability, which she highlighted in ChatGPT's developer mode, to connect servers with powerful write actions.[5]

Before her MCPKit announcement, Lu demonstrated a video-to-Markdown workflow while associated with Frindle. In his meetup recap published by Latent Space on June 1, 2023, Paul Butler described her approach: combine audio transcription and GPT to turn a video into steps, while pulling frames from the recording. The task was to automate the tedious work of keeping documentation up to date.[1]

In her undated essay “Potential applications of GPT”, Lu argued that there was “even more utility in using GPT to grab structure from unstructured text” than in generating unstructured text. Her example came from agency operations: taking information from email conversations and entering it into spreadsheets or web apps, a process she described as tedious and prone to errors or missing data.[3]

Lu also proposed using AI to turn verbal explanations into polished content. She connected this hypothesis to Frindle's practice of interviewing a subject matter expert for half an hour and turning the transcript into a blog post. Translation and rewriting English into idiomatic prose were other applications she advocated; she argued that content should be consumable across languages.[3]

Lu's documentation and content work followed a background in infrastructure engineering and journalism. Her personal site lists previous work in Google Maps infrastructure and Uber observability, alongside technology writing for The New York Times Magazine and New York Magazine. She earned an A.B. in mathematics from Harvard in 2013 and an M.S. in computer science from Columbia in 2015.[2]

In “Building a Technical Content Agency in 2021”, Lu recalled spending about a year exploring startup ideas with her former cofounder David during 2020. After their unsuccessful exploration and amicable separation, she chose to build Frindle outside the venture-backed startup route. She initially fulfilled the work herself, later bringing in part-time contractors for writing, video editing and operations.[4]

Lu focused Frindle on data infrastructure, cloud and DevOps after several early clients came from those areas. Her stated rationale was to accumulate domain expertise and share learning between clients. She also restricted the agency to content containing code or requiring technical knowledge. Documentation testing and updates, technical tutorials, and explainer or demo videos were among the services she developed.[4]

Recruitment changed after an unsuccessful training experiment. Lu ran a two-week course for nontechnical writers, expecting trained journalists to learn to build a Flask webserver; she concluded that she had underestimated prerequisite knowledge, including terminal use and basic programming concepts. She found it easier to recruit technically trained people and teach them to write, drawing on computer science graduate students, new mothers and developers on sabbatical. She offered flexibility and public bylines.[4]