Connecting Claude to edit your docs
Summary
- Once you connect Claude through the Monjiro MCP, you can ask it to create and edit documentation with text instructions.
- Using the documentation site of "WorkingRoom", an open-source project we develop, as an example, we operated Monjiro from Claude Code to create and publish the site.
- From creating the landing page to publishing it took about 5 minutes.
Introduction
This article shows, with a record of actual work, what you can do when you connect Monjiro and Claude Code through MCP. The only things the user does on the Monjiro screen are creating the project and publishing it. All page creation and editing was requested from Claude.
Preparation: Create a project
Before asking Claude to work, create a project in Monjiro to hold the documentation.
- Sign in to Monjiro and click "New project" in the project list.
- Enter the project name in "Title". This time we entered "WorkingRoom".
- In "Source language", choose the language you write the documentation in. This time we chose Japanese.
- Click "Create".
A newly created project has only one page, with the same name as the project. From here on, all work was requested from Claude.
Asking Claude to create a landing page
We entered the following instruction in Claude Code. The only material we gave was the URL of the WorkingRoom GitHub repository.
Using the Monjiro MCP, create a one-page landing page for WorkingRoom, the OSS we develop and publish. https://github.com/kairo-inc/working-room
Claude's report
When the work is done, Claude reports what it did.
The main points of the report were:
- Sources: The page content was written only from the README of the GitHub repository and the documentation in the repository (apps/docs).
- Publishing status: The changes were only saved to the working content and not published yet. It advised us to review them in the Monjiro app before publishing.
- Page structure: The page consists of nine sections: catchphrase, why WorkingRoom, key features (six), architecture diagram, supported AI providers, getting started, roadmap, FAQ and closing.
- Page settings: To make it look like a landing page, it hid the page tree on the left, the breadcrumb and the previous/next page links, and widened the content. It also set the page description shown in search results and on social media.
- Display check: It took screenshots and confirmed that the images, the three-column cards, the diagram and the collapsible sections display correctly.
It also listed two points for a person to check before publishing.
- The "service name and description" used by the AI chat were still empty. This setting cannot be changed through MCP, so it needs to be entered in the Monjiro app.
- The answer to the FAQ "Is it free to use?" was inferred from the license (Apache License 2.0). It needs to be checked against the official policy.
In this way, Claude reports what it could not confirm and what it wrote based on inference instead of leaving it as is. The user only needs to check the listed points before publishing.
The finished landing page
It took about 3 minutes from sending the instruction until Claude finished its report. Here are some sections of the finished landing page.
The finished landing page is published at the link below.
The top of the landing page, with the catchphrase, a product introduction, links to GitHub and the documentation, and an image of the product screen.
The "Key features" section, introducing six features of the product with cards in three columns and two rows.
The "People and AI share the same data space" section. Claude uploaded the concept diagram from the repository and placed it here.
The "FAQ" and closing sections. The questions are collapsible and open their answers when clicked.
Asking Claude to improve the docs from Q&A
Monjiro provides an AI chat by default. So after publishing the landing page, we tried some Q&A, assuming readers asked questions in the chat.
Asking questions as a reader
This time, we asked the following three questions about things the landing page did not cover yet.
- What server specs (CPU and memory) do I need to run WorkingRoom?
- How do I back up WorkingRoom data?
- How do I update when a new version is released?
Monjiro's AI chat answers only from the content of the published documentation. It does not guess about things the documentation does not cover. It honestly answers that the information is not documented and points to related pages.
The question about server specs and the AI chat's answer. It says the information is not documented and shows a related page.
Documentation administrators can see what readers asked in Monjiro's "Q&A history".
The "Q&A history" screen. Readers' questions are listed with the time asked, where they were asked (public site) and the version used to answer.
Claude can also read this history through MCP, so it can find and fill in missing information in the documentation. Next, we asked Claude to improve the documentation based on this history.
Asking Claude for improvements
To improve the documentation, we entered the following instruction in Claude Code. We did not tell it which questions went unanswered, and only asked it to suggest improvements based on the Q&A history.
Using the Monjiro MCP, check the Q&A of the WorkingRoom project and think about what should be added to the documentation.
Claude's suggestions
It took about 1 minute from sending the instruction to getting the suggestions. In that time, Claude read the recent Q&A, compared it with the current page, and checked the facts in the WorkingRoom repository.
The main points of Claude's suggestions were:
- Missing content: For each of the three questions, it identified what the documentation was missing (update steps, what and how to back up, and guidance on required specs).
- Root cause: All three are things people want to know next once they start running WorkingRoom self-hosted. Since the landing page currently ends with "Getting started", it suggested adding a section about operations.
- Drafts checked against the repository: For backups and updates, it read the configuration files in the repository and proposed where the data is stored and concrete steps.
- Priority: It also set priorities: "high" for backups and updates, "medium" for specs.
- No guessing: No official guidance on required specs was found in the repository either. Claude avoided writing guessed numbers, said the development team needs to decide the guidance, and listed only the facts that can be written now.
- Findings beyond the Q&A: It also noticed settings that should be changed when publishing a self-hosted instance, and recommended adding a note about them.
Beyond finding answers to the questions, it worked out what the documentation was missing and how to fill the gaps in about a minute.
Adding an explanation to the FAQ
Following the suggestions, we asked Claude to update the documentation by adding the update steps to the "FAQ" of the landing page. The instruction we entered was:
Please add an explanation of the update steps to the FAQ.
It took about 1 minute from sending the instruction until the addition was done. Claude checked Monjiro's writing guide, read the current page, and then added one new item to the "FAQ". It left the other blocks as they were and only added this item.
The answer to "How do I update when a new version is released?" added by Claude, including the steps, commands and a note.
The added answer includes:
- Four steps: stop, back up, switch to a release tag, then rebuild and start
- The commands to run, and a link to the CHANGELOG where you can check the changes in each version
- Database migrations run automatically at startup
- The database structure may change until v1.0, so always back up before updating
With the documentation changes done, the user can do a final review and publish the page. Once the update is published, Monjiro's AI chat can also read its content and will be able to answer the question about how to update, which it could not answer before.
Summary
This article showed a record of connecting Monjiro and Claude and asking for the following work, starting from an empty project.
| Request | What we gave | Time | Result |
|---|---|---|---|
| Create a landing page | The URL of the GitHub repository | About 3 minutes | Created a landing page with nine sections, including the page settings |
| Review the Q&A | Only an instruction to check the Q&A history | About 1 minute | Suggested missing content with priorities and drafts |
| Add to the FAQ | An instruction on what to add (update steps) | About 1 minute | Added the steps and commands to the documentation's FAQ |
When you connect Claude, you can:
- Create pages just by giving materials: Claude reads your product's repository and materials, plans the page structure, and writes the content, images and page settings.
- Get a report of what to check: Claude reports its sources, what it inferred and what it could not confirm, so you only need to check those points.
- Decide when to publish: Claude's changes only go into the working content and are not published automatically. You can review them in the Monjiro app before publishing.
- Improve docs from readers' questions: Claude reads the Q&A history of the AI chat, finds what the documentation is missing, and goes from suggesting how to fill the gaps to adding the content.
For how to connect Claude, see MCP integration. If you don't use Monjiro yet, you can start from Create an account.
