Takuya Matsuyama's AI-Powered Development Journaling

Table of content

Takuya Matsuyama is a solo developer based in Osaka, Japan. He has been running Inkdrop, a Markdown note-taking app, for over nine years. His YouTube channel devaslife has 200k+ subscribers watching his development workflows. He shares everything on his blog at devas.life.

Takuya built an MCP server for Inkdrop that lets Claude Code automatically write development journals as he works. No manual logging. No interrupting flow state. Just continuous documentation that happens in the background.

Background

Completed a Master’s degree at Nara Institute of Science and Technology (2010)
Built Inkdrop in 2016, now generating $300k+ lifetime revenue
Works as a freelance full-stack developer alongside his indie product
Creates developer workflow content on YouTube since 2018
Father of two, works from home in Osaka

His setup runs on neovim, tmux, and the terminal. Claude Code fits naturally because it operates as a CLI tool.

The Problem with Agentic Speed

When AI agents handle implementation, work moves faster than your brain can track. Takuya noticed he would finish sessions with no clear memory of what got done.

From his blog:

“When you’re in the vibe-coding zone, things progress so fast that you easily forget what you worked on.”

Manual journaling breaks flow. Automated journaling solves this.

The MCP Integration

Takuya built an MCP server that exposes Inkdrop’s note database to Claude. The server is published as an npm package with these tools:

Tool	Function
`read-note`	Retrieve complete note contents by ID
`list-notes`	List notes in a notebook with filtering
`create-note`	Create new notes in specified notebooks
`update-note`	Modify existing note content

Setup requires Inkdrop’s local HTTP server running, then adding the MCP configuration to Claude Code.

The Journaling Workflow

The workflow uses a CLAUDE.md file that instructs Claude to log every task completion. Each journal entry captures:

Task prompt received
What was accomplished
Approach used
Challenges encountered
Suggested improvements

Claude updates the same journal note throughout a session. One continuous record instead of scattered entries.

The Bullet Train Analogy

Takuya compares AI coding to Japan’s Shinkansen: fast for popular destinations but imprecise for specific stops. You still need “local trains” (manual coding) to reach exact requirements.

This means:

Code review becomes the new bottleneck
AI handles execution, humans handle strategy
Boilerplate and migrations are ideal use cases
Novel implementations still need human craftsmanship

His rule: treat Claude as a helper, not a lead developer.

Best Use Cases

Takuya found Claude Code most effective for:

Boilerplate generation — scaffolding and repetitive patterns
Legacy modernization — migrating deprecated dependencies
Small automation — scripts and one-off tools
Tedious refactoring — work nobody wants to write manually

He freed mental capacity by offloading boring tasks. The result: development became fun again.

Key Takeaways

Principle	Implementation
Automate documentation	MCP server writes journals during work
Preserve flow state	No manual logging interrupts
Define clear boundaries	AI executes, humans decide
Match task to tool	Use AI for boilerplate, not novel code
One session, one note	Continuous updates instead of fragments

Getting Started

Install Inkdrop and enable the local HTTP server
Install the MCP server: npm install -g @inkdropapp/mcp-server
Add configuration to ~/.config/claude/claude_desktop_config.json
Create a CLAUDE.md with journaling instructions
Start a session and watch the journal populate

The MCP server source is available on GitHub.