Gemini + Notion API + AWS Lambdaで作る、文脈を理解する英単語学習ツール（MCP対応）This article explains how to build an MCP-compatible English vocabulary…

Reading in a second language often stalls at the same point: you hit an unfamiliar word, paste it into Google Translate or DeepL, get a dictionary gloss, and move on without ever grasping how the word actually functions inside the sentence. A recent Qiita post argues that this word-by-word habit leaves learners with a shallow understanding, and it walks through building a tool designed to close that gap. The project combines Google's Gemini model, the Notion API, and AWS Lambda into a vocabulary-learning workflow that analyzes selected words in their surrounding context and files the results automatically. It is also built to be MCP-compatible, meaning it can be exposed as a Model Context Protocol server.

The core idea is straightforward. Instead of looking up a word in isolation, the user selects a word within a passage of English text, and the tool sends both the word and its sentence to Gemini for analysis. Gemini then returns an interpretation that reflects how the term is used in that specific context rather than a generic definition. The output is written to a Notion database, turning each lookup into a persistent, reviewable record. Over time this builds a personal, context-rich vocabulary log rather than a pile of disconnected translations, which is the differentiating point the author emphasizes.

Each component in the stack plays a distinct role. Gemini, Google's family of large language models, handles the language reasoning: parsing the sentence, inferring the intended meaning, and producing explanations that account for nuance such as idiom, tone, or part of speech. The Notion API serves as the storage and organization layer, letting entries land directly in a structured database that users can sort, tag, and revisit. AWS Lambda provides the serverless compute that ties the pieces together, running the logic on demand without a persistently hosted server. This pay-per-execution model is a common choice for lightweight, event-driven tools of this kind, since it keeps operational overhead and idle costs low.

The MCP compatibility is the detail that situates this project within a broader industry trend. The Model Context Protocol is an open standard introduced by Anthropic in late 2024 to give AI assistants a consistent way to connect with external tools, data sources, and services. Rather than hardcoding a separate integration for every application, developers can expose functionality through an MCP server that any compatible client can call. By making the vocabulary tool MCP-compatible, the author allows it to be invoked from MCP-aware assistants and clients, so the same context-analysis-and-save capability could, in principle, be triggered from within a larger AI workflow rather than only as a standalone app.

That framing matters because MCP has gained notable momentum since its release. A range of clients and platforms have added support for the protocol, and a growing ecosystem of community-built MCP servers now covers tasks from database access to file operations and third-party API calls. A project like this one is a representative example of that pattern: it wraps a practical, narrowly scoped function, in this case contextual word lookup with automatic note-taking, behind an interface that other AI tools can consume. The approach appears well suited to personal productivity and study use cases, where a user wants an AI agent to both explain something and persist the result somewhere durable.

For readers considering something similar, a few prerequisites are worth noting. Using Gemini requires access to Google's API and an associated key, and writing to Notion depends on creating an integration and granting it permission to the target database through the Notion API. Deploying on AWS Lambda involves packaging the code and managing credentials securely, typically alongside services such as API Gateway if an HTTP endpoint is needed. Alternatives exist at each layer: DeepL and Google Translate remain the baseline for quick translation, other LLMs such as OpenAI's or Anthropic's models could perform the same contextual analysis, and databases like Airtable or a simple spreadsheet could stand in for Notion.

Taken together, the article is less a new product announcement than a hands-on integration guide that reflects how MCP is being used in practice. Its central claim, that understanding a word in context is more valuable than an isolated definition, is a familiar principle in language pedagogy, and the technical contribution is showing how off-the-shelf cloud services and a standard protocol can automate that principle. As with any tutorial-style build, the real-world quality of the explanations will depend heavily on Gemini's output for a given sentence, so results are likely to vary by text and prompt design.