Cursor for Teachers: The Complete Guide (September 2026) — Pricing, the 18+ Rule, and Marking Student Code

A Form 3 ICT teacher marks 32 Python submissions every time homework goes out. Running each one, finding the fault, writing a comment — that is an evening gone. Cursor can compress the job into an afternoon. But before anyone clicks subscribe, read one clause: Cursor's terms require users to be at least 18. Your students, as a rule, cannot hold a Cursor account.

This is one of nine tool guides. The full set is in the nine AI tools Hong Kong teachers keep asking about.

What this tool actually is

Cursor is an AI code editor built by Anysphere as a fork of VS Code. It looks and behaves almost identically to VS Code; the difference is that an AI agent sits inside the editor, able to read the whole project, edit across files, run commands in the built-in terminal, and hand every change back to you as a diff to approve line by line.

Let us be blunt: Cursor is a developer tool, not a classroom product. There are no classes, no student rosters, no teacher monitoring view, and nothing connected to curriculum or assessment. It has exactly three honest uses in a school — ICT or computer literacy teachers preparing and demonstrating lessons, teachers building small internal utilities, and marking student programming work. If a school expects it to handle lesson planning, notices or pastoral care, the choice was wrong from the start.

Key features

• Agent mode: describe a goal in plain language and the agent reads files, edits them, runs commands, and presents every change as a reviewable diff.
• Two usage pools: the Cursor Models pool covers first-party Grok 4.6, Grok 4.5 and Composer 2.5; the Other Models pool covers third-party models from OpenAI, Anthropic and Google at their own API rates.
• Composer 2.5: Cursor's own model, released 18 May 2026, built on Moonshot's Kimi K2.5 open-source checkpoint, tuned for sustained long-running work.
• Tab completion: unlimited on paid plans, and the single most visually convincing feature when demonstrating to a class.
• Cloud agents and automations: hand work to the cloud instead of tying up the classroom machine.
• Bugbot code review: agentic review of changes, included on Teams and billed by usage on individual plans.
• Privacy mode: when enabled, Cursor guarantees that neither it nor its model providers train on your code.
• MCP, skills and hooks: connect external tools and data, and run your own commands before or after agent actions.

Full tutorial: marking a Form 3 Python assignment with Cursor

The scenario: Form 3 ICT, unit on lists and loops. You have 32 .py files and a week to return them, each needing a specific comment and a next step.

Step 1 — install and sign up. Download the installer for your operating system from cursor.com. Choose the free Hobby plan at first launch; no credit card is required. The teacher must be 18 or over, which the pricing section below returns to.

Step 2 — set up the folder and de-identify. Create a folder called f3-ict-marking and rename the submissions s01.py through s32.py, keeping the student-number mapping in a file outside the Cursor project. Do not skip this. The AI has no reason to know a child's name, and once a name enters the conversation you have handed personal data to a third party. For the underlying principles see student AI safety: four-layer gating, PII protection and teacher monitoring.

Step 3 — put the marking rubric in the project. Add rubric.md to the folder with your panel's criteria, for example four bands of 5 marks each: correctness, variable naming, comments and readability, and program structure. The agent will cite this file directly, so the comments match your panel's standard rather than a standard the model invented.

Step 4 — open the project and pick a model. Use File > Open Folder to open f3-ict-marking. Press Ctrl+L (Cmd+L on macOS) to open the Agent pane and choose a model from the selector at the bottom. Composer 2.5 is ample for routine marking; there is no need to reach for the most expensive frontier model, and the cost difference is set out below.

Step 5 — write the first prompt. Use @ to reference files so the agent reads exactly what you mean:

@rubric.md @s01.py

You are a Hong Kong secondary ICT teacher. Mark s01.py against the four bands in rubric.md. This is Form 3 homework on lists and loops.

Output:
1. A mark out of 5 for each band with two specific comments each, citing actual line numbers in the student's code.
2. Three things this student should practise next.
3. If the program does not run, identify where it fails but do not write the corrected code.

Constraints: write the comments in Traditional Chinese (Hong Kong usage); keep code and variable names in the original English. Do not modify s01.py. Where you are unsure, write "teacher to verify" rather than guessing what the student intended.

Note the constraints. Without them the agent will cheerfully fix the homework for you, which is not marking — it is doing the work.

Step 6 — run the batch and produce a feedback file. Once the first one looks right, let the agent handle the rest into a single file:

Mark s02.py through s32.py using the same rubric and format, writing all results into feedback.md, one section per student with the filename as the heading.
Add a final section called "Class-wide issues" listing every error pattern that appears three or more times, ordered by frequency.

That last section is often worth more than the 32 comments put together: it tells you exactly what to reteach next lesson.

Step 7 — verify in the built-in terminal. The agent's judgement needs checking. Open the built-in terminal with Ctrl and the backtick key, and actually run the code:

cd f3-ict-marking
python s01.py
python -c "import ast,glob; [ast.parse(open(f).read()) for f in glob.glob('s*.py')]"

The second line parses all 32 files in one pass and stops at the first syntax error, which is far quicker than opening each one.

Step 8 — review before returning. Read feedback.md through. AI comments are usually sound in tone and structure, but on the question of whether a particular student actually understands loops, the model cannot see what you saw in the room. The marks and the comments remain the teacher's responsibility.

Pros and cons

Pros

• Writing and editing code is genuinely first-tier; this is not in dispute.
• As a VS Code fork, teachers with VS Code experience start at near-zero learning cost, keeping most extensions and shortcuts.
• The approve-every-diff design makes it visible what the AI changed, which is useful for showing a class that AI is not magic.
• The free Hobby plan needs no credit card, so a teacher can trial it for a term before spending anything.
• Privacy mode is an explicit commitment not to train on your code.
• First-party Composer 2.5 is far cheaper than third-party frontier models and is good enough for routine work.

Cons

• The terms require users to be 18 or over, so student accounts are not compliant.
• No native Traditional Chinese interface; Cursor's own agent and settings screens are English only.
• No education features whatsoever: no classes, no progress tracking, no monitoring, no curriculum alignment.
• Costs are hard to predict. The seat price is a floor, and consumption sets the real bill.
• Teams and Enterprise add a token surcharge on third-party models that is easy to miss in a quote.
• For a teacher with no coding background it is still an editor rather than a chat box, and that entry barrier is real.

Pricing and fee breakdown

All figures are from the official pricing page and documentation as of September 2026, in US dollars and exclusive of tax.

Model token pricing (per million tokens)

Three hidden costs a school should price in:

First, the seat multiplier. Putting 100 staff on standard Teams seats is US$4,000 a month, US$48,000 a year, over HK$370,000. That buys an editor, not a teaching platform. In practice a school should buy for the handful of ICT teachers who will genuinely use it, not for everybody.

Second, the Cursor Token Rate. On Teams and Enterprise, selecting a third-party model adds US$0.25 per million tokens on top of that model's own API price. It applies to included usage, on-demand usage and even bring-your-own-key usage. Cursor's first-party models (Grok and Composer) are exempt — the platform is, quite deliberately, nudging you towards its own models.

Third, usage overage. Cursor's documentation is admirably direct about this: daily agent users typically total US$60 to US$100 a month, and power users running multiple agents or automations often exceed US$200. Those are per-person figures. A school treating Cursor as a standard tool must budget by consumption, not by headcount alone.

As for student accounts, the answer is no. Clause 2 of the terms requires users to be 18 or the local age of majority, whichever is higher. Even where it is technically possible, bulk-creating accounts for secondary students breaches the terms. On discounts, the official student page says simply that anyone can start free and to watch for event promotions — there is no standing student rate.

Compared with the alternatives

Side by side with Edor.ai

To be clear from the outset: Cursor is far better at writing code than Edor.ai, and we neither intend nor expect to compete with it. If your question is "how do I write this code faster", the answer is Cursor, not us. What follows compares the other questions a school actually asks during procurement.

Why a school still needs a school-based platform

Cursor does one thing to an extremely high standard. The gaps it leaves are, unfortunately, the school day.

There is no student role. An 18-year age floor excludes every secondary student in Hong Kong. An ICT teacher can plan, demonstrate and mark with it, but the student side — who is using it, what they asked, whether they simply copied the answer — is not addressed at all. The Edor.ai tutor gives hints rather than answers by default, with the hint ceiling set per task by the teacher and every conversation recorded for review.

It does not know your school. Your scheme of work, your panel's rubric, your notice format: Cursor knows none of it, so the teacher restates it every time. That is exactly what a school knowledge base solves, as set out in a school knowledge base that makes AI cite your own materials and policies.

Its bill follows consumption, not your financial year. School budgets are annual and answerable to an incorporated management committee. A tool priced per seat plus usage has no brake when a term goes badly. We chose a whole-school annual fee with AI usage included because what a school needs is a number it can put in a budget line.

It will not prepare your grant report. The "智" 啟學教 funding asks for evidence of impact. Edor.ai usage records and the PIE module produce that data directly — see how to use the EDB HK$500,000 AI grant.

Edor.ai is built specifically for Hong Kong schools, Hong Kong teachers and Hong Kong students, and that is where all our effort goes. It does not mean we are better at everything — Cursor writes code better than we do, plainly. But on curriculum alignment, safety gating, per-school pricing, data ownership and grant accountability, no general-purpose developer tool, however excellent, is going to solve the problem for a school.

Takeaway

If your school teaches ICT, if teachers build small utilities, if there is programming homework to mark, Cursor is worth buying — for the few teachers who will genuinely use it, not for all 100 staff. Treating it as the school's AI strategy is the decision that blows up at the mid-year budget review: students cannot use it, the interface has no Chinese, and the bill follows tokens. Keep the two purchases separate and the school gains on both.

Next, read 12 questions to ask before buying a school AI platform, or go back to the nine-tool overview.