Claude Code for Teachers: The Complete Guide (September 2026) — Pricing, the 18+ Policy and Auto-testing Student Code

In July 2026 Anthropic launched Claude for Teachers, free for US K-12 educators. The same official documentation says two other things: there are no student accounts, and users under 18 are prohibited from using Claude, enforced with age assurance services. For a Hong Kong school, those two sentences matter more than the price.

This guide covers what Claude Code actually is in September 2026, the parts a teacher can use, a walkthrough you can follow today, and the usage model that schools routinely underestimate. The full set of nine tools is in the nine AI tools Hong Kong teachers keep asking about.

What this tool actually is

Claude Code is Anthropic's agentic coding tool. It reads your whole repository, edits across files, runs commands, and works directly with git — staging changes, writing commit messages, creating branches and opening pull requests. It runs on five surfaces: the terminal CLI, a VS Code extension, a JetBrains plugin, a desktop app and the web.

To be equally blunt here: this is a developer tool, not a classroom product. No classes, no students, no curriculum alignment, and Anthropic does not pretend otherwise. It built a separate thing for teachers — Claude for Teachers — positioned as a lesson-preparation assistant rather than Claude Code, and available only in the United States.

In a Hong Kong school the sensible uses are three: ICT teachers building testing and marking pipelines, IT coordinators handling data conversion and system maintenance, and turning a weekly data chore into a scheduled task.

Key features

• Terminal agent: run claude in a project directory, describe the goal in plain language, and it plans, edits across files and verifies the result.
• CLAUDE.md project memory: a Markdown file in the project root read at the start of every session, where you record coding standards and panel conventions.
• Pipes and scripting: claude -p accepts piped input, so you can feed it a log or a git diff directly.
• Skills and hooks: package repeated workflows as skills, and run your own commands before or after agent actions, such as formatting after every edit.
• Parallel subagents: split a large task across several agents, with a lead agent coordinating and merging results.
• Scheduled tasks: cloud Routines keep running with your machine off; desktop scheduled tasks run locally with access to your files.
• MCP connectors: attach Google Drive, Jira, Slack or a tool your school built.
• Session portability: move the same session between terminal, desktop app and browser.

Full tutorial: an auto-test harness for Form 6 ICT

The scenario: you set a Python exercise weekly and collect 30 submissions. You want a harness that runs your test cases against every file at once and produces a pass/fail summary, while you still write the comments yourself.

Step 1 — check the plan. Claude Code is included in every paid Claude plan and is not available on Free. For an individual teacher, Pro is the starting point (US$20 a month, or US$17 a month billed annually at US$200 up front).

Step 2 — install. On Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

On macOS, Linux or WSL use curl -fsSL https://claude.ai/install.sh | bash; Windows also supports winget install Anthropic.ClaudeCode. Then start it in your project:

cd f6-ict-autotest
claude

The first run asks you to sign in. If your system has an ANTHROPIC_API_KEY environment variable set, Claude Code bills that key instead of your subscription allowance — teachers rarely want this, and it is usually the explanation for an unexpected charge.

Step 3 — write CLAUDE.md. Create the file in the project root; it loads automatically at the start of every session:

This project is an auto-test harness for Form 6 ICT homework in a Hong Kong secondary school.

Rules:
- Python standard library only; school machines have no extra packages installed.
- Student files live in submissions/ named s01.py to s30.py and contain no real names.
- Never modify anything in submissions/ — read and execute only.
- Cap each program at 5 seconds; treat a timeout as a fail and record it.
- All messages and reports in Traditional Chinese (Hong Kong usage).
- Tests judge output correctness only. No marks, no comments; grading is the teacher's job.

That last rule is deliberate. Automated tests settle matters of fact; the teacher settles matters of judgement. The two should not be blurred.

Step 4 — describe the task.

Create run_tests.py that does the following:

1. Read tests/cases.json, an array where each item has name, stdin and expected_stdout.
2. Run every .py file in submissions/ against each case via subprocess, feeding stdin and comparing output.
3. Ignore trailing whitespace and the final newline when comparing, but keep it case-sensitive.
4. Treat a 5-second timeout as a fail recorded as TIMEOUT; record a crash as ERROR plus the first line of the exception.
5. Write report.csv with columns: filename, cases passed, total cases, names of failed cases, error type.
6. Also print a summary to the terminal: the class pass rate per test case, sorted lowest first.

Start by giving me a sample tests/cases.json with three cases about finding the mean of a list.

Step 5 — review, run, fix. Claude Code shows the files it created. Confirm it has not touched submissions/, then run it:

python run_tests.py

If it fails, paste the error back into the conversation, or pipe it in directly:

python run_tests.py 2>&1 | claude -p "This script is failing. Identify the cause and propose the smallest fix. Do not rewrite the whole file."

Step 6 — watch the usage. Long conversations consume both context and allowance. Use /compact to summarise the session and keep going, or /clear once a task is finished. If you are on an API key, /cost shows what the current session has actually spent.

Step 7 — make it a weekly routine. Once the harness is stable, set it up with /schedule or simply run it after each homework collection. That "lowest pass rate first" summary is your teaching list for the next lesson.

Step 8 — the teacher still reads it. Automated tests answer one question: is the output correct? A brute-force solution that happens to produce the right answer, variables named a, b and c, zero comments — none of that shows up. Marks and comments remain yours.

Pros and cons

Pros

• Chinese writing quality is among the best in this class of tool, so reports and comments read well.
• One subscription covers five surfaces: terminal, VS Code, JetBrains, desktop and web.
• CLAUDE.md and skills let a panel capture its conventions in a file instead of passing them on verbally.
• Deep git integration suits internal tools that need a maintenance history.
• Scheduled cloud routines keep running with the machine off, which fits weekly data chores.
• Team and Enterprise plans do not train on your content by default.

Cons

• The 18-and-over policy rules out students entirely, and Anthropic enforces it with age assurance.
• Claude for Teachers is US-only, so Hong Kong teachers do not qualify.
• Usage shares one pool with web chat, so a big terminal job eats the same allowance you need for writing.
• There is no fixed message count, which makes the real capacity hard to estimate in advance.
• No Traditional Chinese interface.
• Anthropic offers no embedding or speech models, so a school cannot run an entire school platform on this one provider.

Pricing and fee breakdown

Figures from the official pricing page and help centre as of September 2026, in US dollars, exclusive of tax.

Model API pricing (per million tokens)

What a school should watch:

The seat multiplier. Putting 100 staff on Team standard seats at the annual rate is US$2,000 a month, US$24,000 a year, about HK$187,000. That buys a development and writing assistant, not a teaching platform. The rational move is to buy for the two to five teachers who will genuinely use it.

Usage windows are not message counts. Every plan resets on a rolling five-hour window, paid plans add weekly limits, and web, desktop, mobile and Claude Code share one pool. In practice that produces an irritating situation: a teacher who runs a large marking job in the terminal may find they cannot use the web app to draft a parent letter that same afternoon. You can enable usage credits to continue at standard API rates, but that turns a fixed cost into a variable one.

How Enterprise is priced. Anthropic states it as US$20 per seat per month plus usage at API rates, billed annually. The seat fee is low precisely because the real cost is consumption, so a school cannot budget from the seat price alone.

Student accounts: no. Anthropic operates an 18-and-over policy enforced with age assurance, and Claude for Teachers explicitly has no student accounts. Schools cannot create pupil accounts.

Education offers. Claude for Teachers is free for verified US K-12 educators, with at least a full year for those signing up by 30 June 2027; Hong Kong teachers do not qualify. For higher education, Anthropic offers a university-wide plan priced on enquiry, and researchers at academic and nonprofit institutions can get Claude at no cost to start. There is currently no equivalent official offer for Hong Kong primary or secondary schools.

Compared with the alternatives

Side by side with Edor.ai

Stated plainly: Claude Code is much stronger than Edor.ai at writing code, refactoring and understanding large repositories, and we do not compete with it there. Here are the other questions a school asks when buying.

Why a school still needs a school-based platform

Anthropic's posture on education deserves credit: it built a dedicated teacher programme, wrote a K-12 data processing agreement, and excluded students explicitly rather than fudging it. That clarity is precisely what marks out the gaps a school must fill itself.

The student side is closed by design. The 18-and-over rule is a product policy enforced with age assurance, not a setting. A school that wants pupils practising questioning and reasoning under gating, supervision and an audit trail needs a different system. The Edor.ai tutor gives hints rather than answers, with the hint ceiling set per task by the teacher and conversations monitorable in real time.

The geographic limit is real. Claude for Teachers is US-only. That is not Anthropic's failing, but the effect on a Hong Kong school is the same: the same product is free for an American teacher and full price here, without the K-12 terms and data processing agreement attached.

The usage pool assumes an individual user. Five-hour windows, weekly caps, chat and coding sharing one allowance: reasonable for one developer, awkward for a school supporting 100 teachers at once. A whole-school annual fee with AI usage included is our answer to that scenario.

It will not write your grant report. The "智" 啟學教 funding wants evidence of impact, and the Edor.ai PIE module produces it from platform data — see how to use the EDB HK$500,000 AI grant. For how the gating actually works, read student AI safety: four-layer gating, PII protection and teacher monitoring.

Edor.ai is built for Hong Kong schools, Hong Kong teachers and Hong Kong students, and that is where all our effort goes. It does not make us better at everything — Claude writes code better than we do, plainly. But curriculum alignment, safety gating, per-school pricing, data ownership and grant accountability are not problems a developer tool is going to solve for a school.

Takeaway

Claude Code is among the most mature agentic coding tools available and is a sound investment for ICT teachers and IT coordinators. Two things to settle first: pupils cannot use it at all, and its usage shares a pool with your own web chats. Buy it for the handful of teachers who will use it, and leave the student side to a platform with gating built in.

To compare model line-ups and prices, see the Anthropic Claude model page or the full model overview.