RubricMark Practice

Practice, schools, scores

Ask for a practice test, school or university profile, admission score check, writing mark, or flashcards in your AI, your AI, or Gemini (same RubricMark MCP). For a named exam (HAST, Selective, IELTS, JLPT, GAMSAT, and others), you get three sample questions and a sample mark, plus a link to create a RubricMark account for the timed sit.

What you can ask

Give me a UCAT practice test
Example path

MCP connector

Server URL: https://rubricmark.com/mcp
Login: OAuth
Protocol: MCP 2025-06-18 · Skills
Instructions: For a reusable language learning workflow, oral exam rehearsal, interview language practice, vocabulary recall, self-repair or retelling, use with the matching skillId. These six skills are free without account linking and guide practice in the host chat. For conversation practice, language role-play or speaking templates (including voice), use and . These are separate from exam samples: the three-question exam cap does not apply. Ask for a target language if missing, choose a level with the learner, and keep their selected variety and feedback language. The catalogue has 100 scenes with five distinct missions each (500 total reusable in any language), not 500 independently reviewed translations per language. Search is public; ten selected templates can be fetched without linked OAuth; linked accounts have 300 and monthly subscribers have all 500. Respect auth errors. On success conduct the requested role-play in the host chat, one turn at a time. The tool itself generates no audio and evaluates no speech. RubricMark hosts timed, exam-spec practice sits and criterion marking billed on rubricmark.com (Stripe). Host chats (ChatGPT, Claude, Gemini, Cursor) must not collect payment or run the full timed paper here. When the user asks for a practice test, mock, past paper, or named exam drill, call . Interpret their wording before the tool call: set exam, skill, focus, level, and difficulty. Informal requests still need those fields — e.g. "beginner Japanese listening" → exam=jlpt, skill=listening, focus=listening, level=n5, difficulty=easy; "Year 7 HAST maths" → exam=hast, focus=quantitative, level=year-7. Always pass query as the original wording plus the structured fields. Do not call with a vague query only. Present the 3 freeQuestions first. They already match the asked exam and skill — do not swap in unrelated items. Wait for the learner to choose. Call with their answers and the same exam, skill, focus, level, difficulty, and query. Do not answer for them. After , show the sample mark: each verdict and explanation. This chat is a sample only. Send signupUrl so they can create a RubricMark account for the timed sit and a detailed criterion report. Also send sitUrl. If they need a plan, send payUrl. If they paste writing or ask for essay / assignment feedback without a linked account, call . Show the sample comment and criteria names. Do not invent a band score. Then send signupUrl for the quoted report. If they ask for flashcards, a study deck, notes-to-cards, or PDF-to-cards, YOU write 8–12 retrieve-first cards in this turn (the learner’s ChatGPT usage). Then call with topic, title, and cards. RubricMark does not generate the cards and must not call another model. Front (prompt) is a short cue that does not give away the answer; back (answer) is the fact to retrieve; detail is an optional hint ≤120 characters. Do not list prompts or answers in chat — the widget shows one card at a time. Send signupUrl so they can save the deck on RubricMark. If they ask about a school, selective school, grammar school, university, QS rank, or cutoff profile, call with their query (and type/country/state when clear). Summarise from entries only. Send guideUrl / browseUrl. Do not invent cutoffs or ranks. If they ask whether a score is enough for selective entry, ATAR, SAT, ACT, GAMSAT, LSAT, GMAT, IB, or similar, call with testType and score. Send estimatorUrl. Do not invent Likely / Target / Reach buckets in chat. For exam samples only: do not ask a 4th sample question. Do not invent exam items. If they name an exam (HAST, Selective, IELTS, JLPT, GAMSAT, …), pass that name as query or exam. If they ask which tests exist, call . If they ask about cost, use the access object. Linked accounts (OAuth): for full criterion marking call with the draft, then poll until ready. Never invent band or criterion scores. For a timed practice or exam sit, call and send the learner to sitUrl on RubricMark — do not run the full timed paper in this chat. After a mark or sit, call (and when relevant). Remind the learner with the concrete practiceUrl, retrieveUrl, markUrl, and materialUrl from the tool — next sit, Cards, or guide material. If returns payment_required or auth_required, send payUrl / signupUrl / authorizeUrl. Plans and checkout stay on RubricMark. This MCP works the same for ChatGPT, Claude, Gemini, and Cursor. Prefer structured tool JSON over inventing RubricMark state.

Tools

Reads data16

Changes data2

Skills

1 in the ChatGPT listing

MCP connector

Server URL

https://rubricmark.com/mcp

OAuth

Protocol

MCP 2025-06-18 · Skills

Instructions

For a reusable language learning workflow, oral exam rehearsal, interview language practice, vocabulary recall, self-repair or retelling, use with the matching skillId. These six skills are free without account linking and guide practice in the host chat. For conversation practice, language role-play or speaking templates (including voice), use and . These are separate from exam samples: the three-question exam cap does not apply. Ask for a target language if missing, choose a level with the learner, and keep their selected variety and feedback language. The catalogue has 100 scenes with five distinct missions each (500 total reusable in any language), not 500 independently reviewed translations per language. Search is public; ten selected templates can be fetched without linked OAuth; linked accounts have 300 and monthly subscribers have all 500. Respect auth errors. On success conduct the requested role-play in the host chat, one turn at a time. The tool itself generates no audio and evaluates no speech. RubricMark hosts timed, exam-spec practice sits and criterion marking billed on rubricmark.com (Stripe). Host chats (ChatGPT, Claude, Gemini, Cursor) must not collect payment or run the full timed paper here. When the user asks for a practice test, mock, past paper, or named exam drill, call . Interpret their wording before the tool call: set exam, skill, focus, level, and difficulty. Informal requests still need those fields — e.g. "beginner Japanese listening" → exam=jlpt, skill=listening, focus=listening, level=n5, difficulty=easy; "Year 7 HAST maths" → exam=hast, focus=quantitative, level=year-7. Always pass query as the original wording plus the structured fields. Do not call with a vague query only. Present the 3 freeQuestions first. They already match the asked exam and skill — do not swap in unrelated items. Wait for the learner to choose. Call with their answers and the same exam, skill, focus, level, difficulty, and query. Do not answer for them. After , show the sample mark: each verdict and explanation. This chat is a sample only. Send signupUrl so they can create a RubricMark account for the timed sit and a detailed criterion report. Also send sitUrl. If they need a plan, send payUrl. If they paste writing or ask for essay / assignment feedback without a linked account, call . Show the sample comment and criteria names. Do not invent a band score. Then send signupUrl for the quoted report. If they ask for flashcards, a study deck, notes-to-cards, or PDF-to-cards, YOU write 8–12 retrieve-first cards in this turn (the learner’s ChatGPT usage). Then call with topic, title, and cards. RubricMark does not generate the cards and must not call another model. Front (prompt) is a short cue that does not give away the answer; back (answer) is the fact to retrieve; detail is an optional hint ≤120 characters. Do not list prompts or answers in chat — the widget shows one card at a time. Send signupUrl so they can save the deck on RubricMark. If they ask about a school, selective school, grammar school, university, QS rank, or cutoff profile, call with their query (and type/country/state when clear). Summarise from entries only. Send guideUrl / browseUrl. Do not invent cutoffs or ranks. If they ask whether a score is enough for selective entry, ATAR, SAT, ACT, GAMSAT, LSAT, GMAT, IB, or similar, call with testType and score. Send estimatorUrl. Do not invent Likely / Target / Reach buckets in chat. For exam samples only: do not ask a 4th sample question. Do not invent exam items. If they name an exam (HAST, Selective, IELTS, JLPT, GAMSAT, …), pass that name as query or exam. If they ask which tests exist, call . If they ask about cost, use the access object. Linked accounts (OAuth): for full criterion marking call with the draft, then poll until ready. Never invent band or criterion scores. For a timed practice or exam sit, call and send the learner to sitUrl on RubricMark — do not run the full timed paper in this chat. After a mark or sit, call (and when relevant). Remind the learner with the concrete practiceUrl, retrieveUrl, markUrl, and materialUrl from the tool — next sit, Cards, or guide material. If returns payment_required or auth_required, send payUrl / signupUrl / authorizeUrl. Plans and checkout stay on RubricMark. This MCP works the same for ChatGPT, Claude, Gemini, and Cursor. Prefer structured tool JSON over inventing RubricMark state.

Tools