DEV Community: Kakusan

How to Verify Whether a ¥200 Hosting Plan Fits a Company Website

Kakusan — Tue, 11 Aug 2026 09:17:34 +0000

Sakupa's Water plan currently costs ¥200 per month for one long-term website, with 100MB of storage, 2GB of monthly transfer, and 20,000 monthly requests. Those numbers describe hosting limits. They do not include site creation, AI tools, domain registration, email, forms, booking, or payments.

Sakupa is developed by the author. This checklist makes the product claim testable before a company site is published.

1. Inspect the production artifact

For a static Next.js site:

npm run build
du -sh out
find out -type f | wc -l
find out -type f -size +1M -print

Do not upload the project root. Publish only the reviewed static output. Stop if it contains .env files, credentials, customer records, unlicensed source images, or other private material.

2. Measure a representative page

Use browser developer tools or an automated page check to record transferred bytes and request count on a cold load. Repeat on the image-heavy page, not only the home page.

A 1MB page with ten requests suggests roughly 2,000 complete cold loads before either monthly limit becomes relevant. It is only a rough estimate because caching, crawlers, repeat visits, and page variation affect usage.

3. List excluded services

Create separate budget rows for the domain, business email, form processing, booking, payments, AI tools, and maintenance. Record free-tier limits and who receives billing alerts.

4. Define the failure response

The plan does not upgrade automatically. Over-limit sites may receive a notice or service restriction. Decide whether the response is to optimize assets, move a large download elsewhere, upgrade manually, or pause a campaign.

The plan fits when the site is static, small, and operationally allowed to stay within explicit limits. A low price is not a substitute for measuring the build and assigning an owner.

References

Sakupa plans, checked August 11, 2026
Sakupa deployment guide, checked August 11, 2026
Next.js static exports, checked August 11, 2026

Generative AI assisted with structure and editing. A human verified the price, limits, calculations, and publication boundary.

How to Build a Construction Company Website with AI: A Static-Site Checklist

Kakusan — Fri, 07 Aug 2026 00:48:48 +0000

Visitors usually need three answers quickly: what work the company accepts, where it works, and whether its project experience matches their job. A gallery without context cannot answer those questions.

This guide shows a business owner how to prepare the facts, use an AI coding tool to build a static Next.js website, review it, and create a temporary public version for checking. The result is a static company site with service areas, project categories, selected work, and inquiry details. It is not a replacement for a booking system, customer database, or professional review.

Decide what the first version must do

Keep the first release narrow. A visitor should understand the business, decide whether it is relevant, and reach the correct next step. Adding accounts or live records turns a simple company website into a software project with different security and operating responsibilities.

Write down the source of truth for each item before opening the AI tool:

Information	What the owner must supply
Work types	New builds, renovation, repair, commercial work, or other actual services
Service area	Cities or distance limits the company really covers
Projects	Photos, dates, scope, and owner permission
Qualifications	Current licenses and certifications that may be advertised
Inquiry	Information needed before an estimate

Remove secrets, private customer material, internal contracts, unpublished financial information, and high-resolution source assets from the working folder. If a fact is missing, mark it for confirmation; do not ask AI to fill the gap convincingly.

Send this planning prompt first:

I lead a company in the [industry] industry. The company is called [company name] and provides [product or service] to [target customer]. A common customer problem is [problem], and our difference from similar companies is [difference]. I want visitors to [book / request a quote / call / buy]. Do not write code yet. First create a one-page list covering the website goal, target customer, core value, information I must supply, and recommended pages. Do not invent case studies, prices, reviews, or credentials.

Read the answer as an information checklist. Correct the facts before asking for layout or code.

Choose a visual direction without copying the example

The reference below is a local working asset from the Sakupa website guide's prompt toolbox. It is for composition, spacing, typography, and visual hierarchy—not for names, copy, figures, reviews, or business claims.

Swiss Modernism suits factual project information. Strong alignment, numbered sections, and disciplined captions help visitors compare work without turning the page into a visual collage.

Attach the image to the AI tool, then use the exact style direction recorded for this article:

Use the attached reference image only for visual direction. Do not copy company names, text, numbers, testimonials, or claims from it.

Design with Swiss Modernism 2.0. Use: strict grid system (12 columns), Helvetica/Inter fonts, mathematical spacing, asymmetric balance, high contrast, minimal decoration, clean hierarchy, single accent color.

Make the layout readable on mobile and desktop. First turn this direction into a design rule sheet covering color, typography, spacing, buttons, images, and dividers. Wait for my approval before implementation.

Approve the resulting design rules before implementation. Check text size, contrast, mobile behavior, navigation, button labels, and whether the style still fits the business.

Build a static Next.js site in small steps

A static site is a set of files produced in advance. It is a good fit for company information that does not need a server to calculate a private response for each visitor. State that boundary explicitly:

Build this business website with the Next.js App Router and ensure output: 'export' produces static files. Pages, images, navigation, and small browser-side interactions are allowed. Do not use a database, login, Next.js API Routes, Server Actions, request-time rendering, private environment variables, or features that require server processing. Before implementation, list any requirement that cannot fit a static website and propose an alternative that needs no backend.

Create service, project, company, and contact sections. Give every project image a useful caption and size images for the web before adding more pages.

Implement only the home page first. Do not create all pages at once. The home page must include a headline that explains the business, a short description, one primary action, three core benefits, genuine trust information, and footer contact details. Reuse the approved design rules. When complete, tell me which devices and content I should check, and wait for my approval before building the next page.

Open the local result yourself. Follow every navigation link, read the page on a phone-sized screen, and replace every placeholder or unconfirmed statement with either an approved fact or a deliberate omission.

Separate automated checks from owner approval

AI can find broken links, missing alternative text, layout problems, and files that should not enter the build. It cannot authorize a photograph, certify a qualification, or decide that a commercial promise is safe.

Perform a pre-delivery review of the current static Next.js website. Separate the result into “Checked by AI” and “I must confirm.” AI should check widths 375, 768, 1024, and 1440; keyboard navigation; text contrast; image alternative text; reduced motion; links; 404 behavior; static build; and whether .env files, secrets, customer data, or non-public files entered the output. The human section must list company information, prices, business promises, permissions, privacy, and form destinations.

For this construction company, the owner must personally approve photo ownership, project addresses, license wording, service boundaries, estimate conditions, and safety claims. Keep a dated copy of the approved text and assets so the source of each public claim is clear.

Install Sakupa MCP, then fully restart the AI tool

For the preview-and-inspection example, this guide uses Sakupa MCP, which is developed by the author. It is one implementation option in the workflow, not a requirement for building the website.

Paste the current configuration into an AI tool that supports MCP:

Install and enable the latest version of Sakupa MCP in the current AI tool using the following configuration:

{
  "mcpServers": {
    "sakupa": {
      "command": "npx",
      "args": ["-y", "@sakupa/mcp@latest"]
    }
  }
}

After saving the configuration, fully quit the AI application and start it again. Merely closing the project, opening another window, or continuing the same chat is not enough. Reopen the website project after the restart.

Now verify the connection before allowing any initialization, inspection, upload, or publication:

Confirm that Sakupa MCP is connected and that the init, analyze, and deploy tools are available. Do not initialize, analyze, upload, or publish anything yet. If the tools are unavailable, check whether the Sakupa MCP configuration was loaded after the AI tool restarted.

Stop here if init, analyze, and deploy are not available. Check the MCP configuration and restart the AI tool again; do not continue on the assumption that the connection probably worked.

Initialize and inspect before publishing

Once the three tools are visible, prepare the open project without publishing it:

Use Sakupa to prepare the project currently open in this AI tool for first-time use. Do not ask me to choose a folder, and do not publish the website. Tell me whether it was successful when complete.

Only after that succeeds, inspect the actual publishable output:

Use Sakupa to inspect the current project and identify the actual website content that can be published. Inspect only; do not upload or make anything public. Tell me what is ready to publish, whether any files should not be made public, and what still requires my confirmation.

Read the inspection result. Remove secrets, private files, source material, and anything you cannot explain. Compare the publishable page with the approved facts, not just with the local preview.

Require a final confirmation before deployment

Publishing changes who can see the site. The AI tool must explain access, duration, and anything the owner needs to retain, then wait for an explicit answer:

Use Sakupa to publish the website that passed the previous inspection. Before making it public for the first time, explain who can access it, how long the free website remains available, and what I need to keep safe. Wait for my explicit confirmation before publishing. When complete, give me the working URL and the expiration time with time zone.

Open the returned URL in a separate browser session. Check the title, images, mobile layout, links, contact destination, and expiration time. A successful tool response is not proof that the public page is correct.

Keep private or dynamic work outside the static site

Do not place live project dashboards, subcontractor systems, customer records, and estimate calculators that require private data in this public build. Use an established service designed for those jobs and link to it when appropriate. This keeps the company site easy to review and reduces the amount of sensitive data handled by the website project.

Construction licensing, advertising, warranty, and safety rules differ across countries and regions. Have a qualified local reviewer check regulated statements.

The practical stopping point is a small site whose facts the owner can verify and whose public output has been inspected. Add another feature only when someone has accepted responsibility for its data, security, cost, and maintenance.

References

Your MCP Tool Is Installed but Missing: Restart Before You Deploy

Kakusan — Mon, 03 Aug 2026 15:03:20 +0000

Your MCP Tool Is Installed but Missing: Restart Before You Deploy

Adding an MCP server to an AI tool's configuration does not load it into a process that is already running. If a tutorial jumps directly from configuration to deployment, a non-technical user may continue with a tool that is not actually connected.

This guide builds a small logistics-company site as static files, then uses an explicit restart and verification gate before any public action.

Keep the business site separate from operations

I operate a small logistics company. Organize my approved facts into services, supported cargo, service area, opening hours, process, exclusions, and contact route. Keep dispatch, tracking, contracts, driver records, customer data, quotes, and payments outside the public website. Mark missing facts as UNKNOWN. Do not invent delivery times, prices, fleet size, certifications, customers, reviews, or guarantees.

Create the visual rules before code

Use the attached reference image only for visual direction. Do not copy company names, text, numbers, testimonials, or claims from it. Create a functional logistics-company design rule sheet with clear service categories, area and cargo conditions, high-contrast contact actions, responsive spacing, and restrained motion. Wait for approval before implementation.

Export static files

Build the approved website with the Next.js App Router and output: 'export'. Do not add tracking, dispatch, accounts, contracts, quotes, payment, customer data, a database, API Routes, Server Actions, request-time rendering, or private environment variables. Link operational functions to approved external systems and list every non-static requirement before coding.

Run the production build and inspect the generated output. Confirm the routes, images, contact link, mobile layout, and absence of UNKNOWN before adding a publishing tool.

Install the latest Sakupa MCP

The publishing example uses Sakupa MCP, which is developed by the author. Add the current package configuration to the MCP-capable AI tool:

{
  "mcpServers": {
    "sakupa": {
      "command": "npx",
      "args": ["-y", "@sakupa/mcp@latest"]
    }
  }
}

Fully quit and restart the AI tool

This is a required step. Quit the AI application completely, start it again, and reopen the website project. Opening another chat or window inside the same running application does not prove that the new MCP configuration was loaded.

After the restart, verify the connection without initializing or publishing:

Confirm that Sakupa MCP is connected and that the init, analyze, and deploy tools are available. Do not initialize, analyze, upload, or publish anything yet. If the tools are unavailable, check whether the configuration was loaded after the AI tool restarted.

If init, analyze, or deploy is missing, stop. Check the configuration and restart again instead of assuming installation succeeded.

Initialize without publishing

Use Sakupa to prepare the project I currently have open for first use. Do not ask me to choose a folder, and do not publish the website. Tell me whether preparation succeeded.

Initialization prepares local project management. It does not make the website public.

Analyze without uploading

Use Sakupa to inspect the current project and identify the actual publishable website output. Inspect only; do not upload or make anything public. Tell me which directory would be published, whether it contains files that must stay private, and what still needs my confirmation.

The selected output must not contain credentials, .env files, customer data, contracts, dispatch records, or source material that was not approved for publication.

Deploy only after explicit confirmation

Use Sakupa to publish the website that passed inspection. Before the first public action, explain who can access it, how long the free website remains available, and which management credential must stay private. Wait for my explicit confirmation before publishing. After publication, return the public URL and the expiration time with its timezone.

The current manual says the free website is public for 24 hours. .sakupa/site.json is a management credential and must not be committed to a public repository.

Verify the public result

Open the returned URL in a signed-out browser.
Confirm the title, routes, images, and contact link.
Check that no operational or customer data is visible.
Record the URL, publication time, and timezone.
Recheck availability after the platform's required review interval.

The restart gate is deliberately boring. It separates “configuration was saved” from “the tool is available,” which is exactly the distinction a safe deployment workflow needs.

References

Generative AI was used for structure and editing. The MCP sequence and current manual were checked on August 3, 2026.