On Monday morning, this person no longer opens Craft. In the past, they would launch the app, duplicate a template, change the date, and arrange the week's tasks. Now they simply tell Claude, "Make this week's version with the usual template." Seconds later, a well-structured document is waiting inside Craft. One action so ordinary it had gone unnoticed, opening the app's screen, has disappeared.
Behind this small change, Craft is quietly transforming from a "beautiful note-taking app" into a "knowledge hub that works with AI." Let us follow, step by step, what made that possible.
A Design Philosophy That “Dislikes Friction”
The first point to understand is that Craft was conceived around an intense dislike of friction. It is a native app with an especially high degree of polish on macOS, iOS, and iPadOS, with Windows, web, and Android versions also available.
Unlike many tools built on the web, it is optimized as a native application, so there is almost no wait for loading. The speed with which a thought can become text removes the first source of friction.
Craft also takes an "offline-first" approach, keeping data locally on the device and automatically syncing it when the connection returns. Work does not stop while traveling or in places with a weak signal, and knowing the data is at hand brings reassurance.
Typefaces and spacing are carefully arranged so that a written document is immediately good enough to publish. Craft deliberately avoids a full relational database for the same reason: it lets users concentrate on writing and thinking without being distracted by "building a system" through property design and view construction. A Collections feature allows simple aggregation, but the writing experience remains the principal concern.
The “AI Wall” Standing Before Accumulated Knowledge
However pleasant writing may be, a different wall appears when you try to use the knowledge you have accumulated: the AI wall. Large language models such as Claude are astonishingly intelligent, yet left alone they know nothing about the content of your notes.
Until now, letting AI read local documents required a dedicated integration for every app. Code multiplied with the number of combinations, making the approach highly impractical. What broke through this wall was the Model Context Protocol, or MCP, proposed by Anthropic in late 2024. It is often called "USB-C for AI."
By standardizing communication between an AI app, the host, and the data side, the server, MCP allows them to exchange information safely through one configuration without a separate implementation for every pairing. It separates the roles of the host interacting with the user, the client mediating communication, and the server providing data. One Claude instance can therefore connect simultaneously to several servers, including Craft, a file system, and GitHub.
Craft made itself available as such a server and gave external AI permission to read and write notes within a selected space. At this point, the "beautiful notes" become "knowledge within AI's reach."
Once Connected, Everyday Work Looks Different
To connect them, select the space you want to share in Craft and create an MCP connection, then register that URL in Claude. The only remaining step is to approve access on the authorization screen that opens in a browser the first time.
Imagine recording an idea as a voice memo in a taxi. After the app automatically summarizes and transcribes it, ask Claude, "Retrieve the latest summary and add it to the relevant block in this week's Craft document." Because Craft gives its content a block-level structure, Claude can understand exactly what to add and where and carry out the request.
With multiple servers connected to one Claude instance, Claude acts as a bridge for information. It retrieves from one side, organizes the meaning, and writes to the other. A process that once required complex conditional branches in an integration tool now runs from a single natural-language instruction.
Capture Roughly, Let AI Organize
A more advanced use reveals the idea underneath this approach. Put every thought into a receptacle in Craft without classifying or organizing it. Later, leave the task to Claude by saying, "Read what has accumulated and distribute it to the appropriate pages according to context."
The most troublesome cost in information management was deciding where to put something at the moment it was saved. When people hesitate over classification, they stop saving at all. This approach separates recording from organizing and enables a division of roles: capture roughly, let AI organize.
At the end of this progression, we return to the opening scene. Even the last friction of opening an app and searching for a template has disappeared. People no longer operate a screen. They only communicate their intention in words, and the work is completed beyond the screen. Here lies a quiet transition from an age of "operating" software to one of "handing intentions" to AI.
Notion Closes; Craft Opens
Craft's position becomes clear when compared with its greatest point of reference, Notion. Though they look similar, their philosophies are almost opposites. Armed with diverse properties, tables, Kanban boards, and calendars, Notion is like a castle for building a complex management system within one place.
Its approach to AI reflects that idea. Notion keeps its own AI within the system and answers across an entire workspace. As of 2026, full AI has shifted into the upper-tier Business plan at $20 per month, and the former standalone AI add-on has been discontinued. The direction is toward completing everything inside itself.
Craft, by contrast, chose to open a window outward. Through MCP it presents its data to current external models such as Claude and Cursor, combining the continually developing reasoning power of AI with its own lightweight data foundation as needed. Notion closes and encloses; Craft opens and connects. This contrast clearly expresses their characters.
Do Not Entrust Everything to One Tool
For this reason, the age of covering everything with one tool is beginning to pass. The more universal a tool becomes, the more work it asks users to spend building and maintaining the system, and the more time its management tends to consume. An alternative idea is spreading: identify each tool's strengths, combine them, and use AI as the bridge. For example, Obsidian, with complete local storage, long-term Markdown data ownership, and more than 2,500 community plugins, can serve as the knowledge mothership; Notion can handle complex management; and Craft can provide intuitive notes and sharing and act as the window for dialogue with AI.
The market also includes tools with distinct characters: Fabric for interpreting every kind of media, Reflect with its emphasis on encryption, Mem for organizing notes automatically, Granola for meeting records, and Apple Notes with deep Apple integration. Among them, Craft stands precisely between the ease of Apple Notes and advanced processing by state-of-the-art external AI.
Hand Over Intention in Words
Craft's role here resembles that of an interpreter connecting people and AI. Instead of having AI read raw data or complex tables directly, place the information into the orderly vessel of Craft first and then pass it along. This produces more precise answers. The features that structure documents visually do more than lighten the burden on people; they also function as an excellent format for AI to interpret information correctly.
Like the person who stopped opening a screen on Monday mornings, future ways of working may begin by putting intention into words and handing it over. Craft is quietly establishing a firm place as a destination for that handoff.
