An MCP connector for your AI (best with Claude or Codex)

A senior ELT editor's eye,
inside your AI.

AI can write an ELT lesson in seconds. But its quality is wildly variable, and you're entirely in the hands of whatever training data your AI calls on. And the chances of hallucination are high.

ELT Brain equips your AI with the knowledge and tools it needs to reliably design tasks, vocab sets, test items and lessons that hold up to scrutiny, and to edit the way a senior commissioning editor would – consistently, with SLA-cited reasoning behind every decision.

Grounded in second-language acquisition research, not vibes
25
Tools (9 learning science + 16 ELT-specific)
333
Research notes
6
Expert personas
2mins
To connect
claude · with elt-brain · mcp
connected
you
Review this B1 dialogue and the comprehension exercise. Suggest fixes for anything that wouldn't pass an experienced ELT content editor.
claude
Running the input-quality audit, then I'll regenerate the weakest comprehension item.
→ doctor_audit_for_input_quality()
claimed level B1 · auditing 6 categories · 00:00:04
⚠ Q2 labelled "reading for gist" – actually tests specific dates
⚠ Dialogue forces present perfect into an exchange that wouldn't use it
⚠ Two MCQ distractors equally defensible from the text
→ write_reading_task(rewrite=q2)
rewriting Q2 as specific-info item · adding text-evidence
✓ Item rewritten: matches construct, distractors now catch real misreadings
✓ Citations: Alderson (2000), Khalifa & Weir (2009)
→ lb_cite_sources()
surfacing the research behind each flag
✓ 8 source notes attached – naturalness, construct validity, level coherence
Three substantive flags, all fixed and cited. The dialogue needs a small rewrite at line 4 to drop the forced present perfect – want me to draft that?
Yes, and run the vocab set against the level
§ 01 THE PROBLEM

AI writes ELT content instantly. But it often gets it wrong.

AI can now write an ELT lesson, a reading task, or a vocabulary set in seconds – and that's exactly the problem. It produces content that looks good at first glance, but falls apart on closer inspection: a "reading for gist" task that actually tests specific dates, a B1 text seeded with unglossed C1 collocations, a vocabulary set with no frequency rationale and no recycling plan, a dialogue that forces the present perfect where no fluent speaker would use it.

Catching it is the time-consuming work that an ELT pro does – work that's hard to scale and expensive in both time and money.

There's more material to check than ever, and the same number of hours to check it in. So flaws creep through: into production, where they cost a re-edit or a reprint, or into the classroom, where they result in a lesson that doesn't work with your group.

ELT Brain applies a senior editorial eye automatically – at the brief, at the writing desk, in the lesson prep, and at the editorial pass – with cited reasoning from learning science and SLA research at every step, so your time can be focused on the judgement only a human can apply.

§ 02 THE PROOF

See the difference ELT Brain makes.

We asked an AI to draft a B1 reading task and a vocabulary set for the same source text, with and without ELT Brain.

A
A B1 reading task, without vs. with
+

Without

5 comprehension questions labelled "general understanding." Two are MCQs whose distractors say almost nothing about misreadings. One is a "T/F" item where neither answer is reliably supported by the text. The level looks B1 by feel, with a couple of C1 collocations sitting unflagged.

With

3 items, each with a stated sub-skill (gist, specific-info, inference), a construct that matches the form, distractors with a stated rationale for each misreading they catch, and the text-evidence span quoted underneath. Lexical coverage assessed at 94% against the Oxford 3000 / NGSL – one C1 item flagged with a glossing suggestion.

B
A vocabulary set, without vs. with
+

Without

15 words pulled from the text, presented as word + translation pairs. No frequency band stated, no collocations, no receptive/productive split. No recycling plan. The selection logic, if you ask, is "these came up in the text."

With

10 items selected on a stated basis (frequency-priority, with two text-specific stretch items flagged). Each item carries word, lemma, POS, IPA, frequency band (cited to NGSL), 3 most-useful collocations, an example sentence at level, and a receptive/productive target. A recycling plan over the next three lessons.

C
A coursebook-page audit, without vs. with
+

Without

"The dialogue is engaging and the activities are well-designed. Consider adding more interactive elements." Three lines of bland reassurance and no specific flags.

With

Six substantive flags: forced present-perfect at line 4, MCQ distractors equally defensible from the text in Q2, vocabulary item pretending to be B1 when it's solidly C1, "reading for gist" question testing specific information, gender-coded character defaults, no spaced-recycling plan. Each flag has a location, severity, cited reasoning, and a concrete suggested rewrite.

Start using it →
§ 03 THE TEAM

ELT Brain gives you six specialists, always on call.

Your AI picks the right specialist automatically, based on your prompt – or you can call one by name.

Connect ELT Brain to the AI you already use: Claude, ChatGPT, Codex, Cursor, or anything that supports MCP connectors and Skills.

Once you have ELT Brain connected, your ELT-related prompts are routed through the team of six specialists: the Learning Scientist, the Brief & Context elicitor, the Curriculum Architect, the Instructional Writer, the Course Doctor, and the Delivery Coach.

↓ Click each specialist to see them in action
01
Learning Scientist
the evidence-checker
ls_*
4 tools

Tells you what the SLA research actually says, with cited sources and evidence strength. Settles design arguments like explicit vs implicit grammar, PPP vs TBLT, coursebook vs emergent curriculum without fence-sitting. Flags popular ELT myths the moment they show up.

"You're claiming learners need to be exposed to the target structure 'lots of times' to acquire it. The research is sharper than that – frequency matters, but so does noticing and pushed output (Schmidt 1990; Swain 1995). Mere exposure without noticing produces input that isn't intake. Here's what to do about that in this lesson…"
Evidence lookupPrinciple explanationSymptom diagnosisTension resolution
02
Brief & Context
the elicitor
lb_*
5 tools

Pushes you to provide the inputs ELT design actually needs: CEFR level, L1, target skill focus, instructional setting (EFL / ESL / EAP / ESP / business / young learners / exam prep), exam target if any. Pushes back on briefs that don't add up. Pulls relevant case studies when a principle needs concrete grounding.

"The brief says 'design a lesson on speaking confidence for adult learners.' That's not a brief, it's a wish. Speaking confidence at A1 looks completely different from B2. I need: CEFR level, L1 group if it matters for transfer, instructional setting, and what 'confidence' actually means here – willingness to communicate, fluency at speed, or accuracy under pressure. Then I'll design it."
Learner contextCourse briefBrief pushbackWorked examplesSource citations
03
Curriculum Architect
the lesson designer
arch_*
2 ELT tools

Produces single-lesson skeletons and language syllabi with coherent progression. Lesson stages have timed durations, predicted comprehension and production demand, and a feedback slot designed in. Syllabi are tied to communicative performance goals, CEFR/ACTFL level, recycling, assessment evidence, and teacher workload. Refuses content-shaped goals ("cover present perfect") and aspirational ones ("become more fluent").

"Your goal is 'cover the past simple.' That's content, not an outcome. A 60-minute lesson can move learners from declarative knowledge into early procedural use – not to mastery. Try: 'By the end, learners can narrate a past event in 4–6 sentences on a familiar topic, using past simple with time markers.' Now I can design the stages."
ELT lesson designLanguage syllabus design
04
Instructional Writer
the materials craftsperson
write_*
9 ELT tools

Writes the materials that actually go on a page. Vocabulary sets with selection rationale, full per-item depth, and a spaced-recycling plan. Reading and listening tasks where the item-type matches the construct. Graded grammar practice that moves controlled → contextualised → freer, with the expected error profile attached. Speaking tasks that have a real outcome and a real gap. Also writes language clarifications, performance evaluations, corrective feedback, and supported Cambridge English, Oxford Test of English, and IELTS exam items.

"You asked for a 20-item vocabulary set for a single B1 lesson. Working-memory and retention research support 6–12 items per single lesson; above 15 you're producing reference material, not a teachable set. Either reduce to 10 with a recycling plan over the next three lessons, or reframe as a unit-level vocabulary focus – which is fine, but it needs to look like one."
Vocabulary setsReading tasksListening tasksGrammar practiceSpeaking tasksLanguage clarificationPerformance evaluationCorrective feedbackExam items
05
Course Doctor
the editorial auditor
doctor_*
2 ELT tools

Reads a reading text, listening script, dialogue model, exercise, full coursebook page, or supported exam item and runs the audit a senior commissioning editor or assessment specialist would: naturalness, CEFR level coherence, construct validity, hidden cultural and contextual assumptions, inclusion, teach-vs-test mismatch, pedagogical coherence, and exam-spec compliance. Returns flags with location, severity, cited reasoning, and concrete suggested revisions.

"This is a polished dialogue with three substantive problems. Line 4 forces present perfect into an exchange a fluent speaker wouldn't use it in (the past simple is the natural form here). Q3 is labelled 'reading for gist' but tests recall of a specific date in paragraph 2. The character roster is six men and one woman, all from the same cultural context. Here's the rewrite, the construct-corrected item, and a revised cast."
Input-quality auditExam item audit
06
Delivery Coach
the session designer
coach_*
3 ELT tools

Designs synchronous speaking sessions, classroom activity flow, and learner-affect support. It matches interaction patterns to class size and mode, plans setup and transitions, names monitoring focus per stage, and supports anxiety / willingness-to-communicate barriers without lowering the language demand into mere comfort.

"You're running a 75-minute live online session on negotiation for 14 B2 learners. Don't open with 'discuss in groups' – that's the most common failure mode for online speaking sessions. Open with a structured warmer that gets each pair talking individually first, then introduce the task. Here's the run sheet with CCQs, monitoring focus, and a feedback slot designed around the cognitive load of the final ranking task."
Speaking session designClassroom activity managementLearner-affect support

Behind the team sits a curated knowledge base: 333 research notes drawn from the SLA and materials-design literature, plus the cross-cutting learning-science research. Every tool cites its sources.

§ 04 IN USE

How you could use it

01

Draft a B1 reading lesson with matched taskLesson skeleton, vocabulary set, and reading task in one flow – all level-coherent, all based on cited SLA research.

"Design a 60-minute B1 reading lesson on starting a new job, with a 10-item vocabulary set and a 4-item reading task"
02

Audit a freelancer's draft before sign-offThe red-pen pass a senior editor does by hand, applied consistently in seconds.

"Audit this B1 coursebook page for naturalness, level coherence, and construct validity: [paste]"
03

Adapt a coursebook lesson for your groupTomorrow's page customised to mixed L1s, a specific exam target, or a level that's slightly off – the night before, not the night after.

"Adapt this B1 coursebook page for my class of Spanish-L1 adults preparing for FCE: [paste]"
04

Defend a design choice with cited evidenceWhen an editor or trainer pushes back, name the research finding behind your decision.

"My editor wants more explicit grammar drilling. What does the SLA research say about timing of focus-on-form at B1?"
05

Design a speaking session for a real class sizeLesson plan with interaction patterns matched to mode, CCQs, monitoring focus, and a feedback slot designed in.

"Design a 75-minute live online speaking session on negotiation for 14 B2 learners"
06

Write a listening task that respects connected speechBottom-up priming named explicitly. Item-types matched to listen-pass. No A2 + natural-fast.

"Write a B1 listening task on this radio interview, 4 items, at natural-slow speech rate"
07

Build a CEFR-aligned speaking rating descriptorFor formative or summative use, anchored to intelligibility rather than vague impressions.

"Write a B2 ranking-and-decision speaking task with a rating descriptor for formative assessment"
08

Train new editors against a consistent rubricThe same flags the senior editor would catch, every time, with the citations behind them.

"Audit this trainee's draft and explain each flag so I can use the audit in supervision"
§ 05 AUDIENCE

Who it's for

01

ELT writers (freelance or in-house)

You're producing material to a publisher's brief, often under time pressure. The lesson skeletons, vocabulary sets, reading and listening tasks all run a self-check against an SLA-informed rubric before they come back to you. Things that would get picked up by an editor or reviewer (mis-pitched lexical load, aspirational goals, no recycling plan, construct-invalid items) get caught before you send anything.

And every output cites the research behind its decisions, so if your editor asks "why?" you have an answer that isn't "this felt right."

02

ELT editors

You're reviewing freelancer drafts at scale, and consistency is hard. One editor can only read so many pages a week. The audit tool runs the editorial pass you'd do manually, picking up issues and improvements, with location, severity, cited reasoning, and a concrete suggested rewrite.

The result: every page gets a consistent editorial pass; editors can then apply their judgment to the things ELT Brain has highlighted. The time required per editorial pass is dramatically reduced.

03

Teacher-trainers

CELTA, DELTA, CertTESOL trainees tend to produce intuition-led lesson plans. Every ELT Brain output carries an explicit research trail, so trainees can read the cited findings and trace the reasoning behind each design choice.

Useful in supervision too: when a trainee writes a content-shaped goal, the tool refuses it and explains why, before you have to.

04

Classroom teachers

You teach a specific group with specific needs – maybe mixed L1s, a particular exam target, an awkward textbook page that doesn't quite fit, or an evening lesson with twelve learners at three different levels. Every week you adapt material, write extra tasks, or design supplementary work, often the night before. ELT Brain is a huge time-saver.

Hand it tomorrow's coursebook page plus your group's profile and ask for a customised version. Ask it to write a warmer and three extra tasks at the right level. Ask it to audit what you're about to teach and flag what won't land with this group. You have the expertise of a senior trainer available whenever you need it.

§ 06 RIGOUR

Grounded in the SLA and materials-design literature. Designed to push back, not please.

Every design decision ELT Brain makes can be traced to a specific research finding. The knowledge base covers 15 ELT-specific domains plus universal learning-science research, and every finding carries an evidence-strength rating: moderate evidence is never laundered as strong. When the research is silent on a topic, the tools say so rather than fall back to general AI guesswork (which is what your AI tool is likely to do without ELT Brain).

AI models are trained to be agreeable. They soften criticism, pass material that should fail, and tell you what you want to hear. ELT Brain pushes back on that in three ways.

01

Cited research, never invention.

Every citation comes from the ELT/SLA knowledge base, so you won't get fabricated references. Popular ELT myths ("younger always means faster," "native speakers are always better teachers," learning styles, the critical-period as a hard cliff) are flagged the moment they appear in a request.

02

Some requests get refused.

Content-focused lesson goals ("cover present perfect"). Aspirational ones ("become more fluent"). A vocabulary set of 30 items for a single lesson. A "reading for gist" task with 8 gist items. A natural-fast listening task for A2 learners. Each refusal explains why and tells you what you should do instead.

03

Every output is self-reviewed.

ELT Brain checks its own work before presenting anything back to you. Fifteen ELT-specific rubrics tell the tools what the work must answer to before it comes back. Vague goals are sharpened before you see them. Mis-pitched vocabulary gets caught. Construct-invalid items get rewritten. Audits with no cited reasoning per flag are rejected.

Two layers of evidence

ELT Brain sits on two layers of research. The foundation is general learning science – how people learn anything: cognitive load, retrieval, transfer, multimedia, assessment, instructional design, learner psychology. On top sits a deep ELT layer: SLA foundations (Krashen, Long, Swain, Schmidt, DeKeyser, Lantolf, Larsen-Freeman, N. Ellis), the four skills, language testing (Bachman-Palmer validity, ALTE standards, fairness and bias), the CEFR Companion Volume 2020 first-class additions plus ACTFL Proficiency Guidelines and the OPI, vocabulary acquisition (including lexical bundles), grammar instruction, discourse and pragmatics, sociolinguistic context (World Englishes, ELF, Schneider's Dynamic Model, translanguaging, native-speakerism, critical applied linguistics, MTB-MLE), specific contexts (CLIL, EMI, EAL, ESOL, young learners with story-based teaching, exam prep, business, EAP including BALEAP frameworks, ESP, BICS/CALP and Cummins), methodology and materials (including TPR), inclusion and accessibility (dyslexia, neurodiversity, refugee/migrant), and the publisher-side disciplines (scope-and-sequence, materials development, image and representation).

The foundation is publisher-neutral. It draws on Oxford 3000/5000, Cambridge English Profile, NGSL, AWL, ALTE standards, CEFR descriptors, and the wider applied-linguistics literature – without privileging any single publisher's output.

§ 07 QUESTIONS

Common questions

01
What is this, exactly?

ELT Brain is a service your AI talks to when you ask it for ELT design or audit help. It supplies the SLA and materials-design expertise that general-purpose AI models don't have on their own. You still use your AI the way you usually do – ELT Brain just makes what comes back structurally sound, level-coherent, and defensible against the rubric a senior commissioning editor would apply.

It produces design artefacts: lesson skeletons, vocabulary sets, reading and listening tasks, grammar practice, speaking tasks, audit reports. It doesn't produce finished coursebook PDFs or video content – it makes the design sound so the production work doesn't get wasted.

+
02
Do I need to be technical to use this?

No. You just need to be able to install a connector and a Skill in your AI tool – don't worry if you don't know what that means: simple step-by-step instructions are provided, and it takes 1-2 minutes.

+
03
Which AI tools does it work with?

ELT Brain works best with Claude or OpenAI's Codex. You'll need a paid plan. It also works with any other AI that supports MCP connectors and custom Skills. You can use it with ChatGPT, but if you have a paid ChatGPT plan, you have access to Codex which supports ELT Brain much better.

+
04
Does it replace my AI, or work with it?

It works with the AI you already use. ELT Brain supplies the expertise; your AI supplies the writing. The simplest way to think of it is that ELT Brain supplements your own prompts automatically, providing your AI with ELT expertise in order to guide its responses.

+
05
What's the knowledge base built on?

ELT Brain is based on 333 hand-curated research notes, each written as a complete argument with primary citations and an evidence-strength rating:

  • 195 'core' learning science notes covering cognitive load, retrieval practice, assessment, and instructional design.
  • 138 ELT-specific notes span 15 domains: SLA foundations (including sociocultural SLA, CDST, and usage-based SLA), vocabulary (including lexical bundles), grammar, the four skills, language testing, methodology and materials (including TPR), learner factors, sociolinguistic context (including Schneider's Dynamic Model and critical applied linguistics), CEFR and frameworks (including ACTFL), materials and publisher workflows, digital and AI in ELT, inclusion and accessibility, teacher development, discourse and pragmatics, and specific contexts (CLIL, EMI, EAL, ESOL, business, EAP including BALEAP frameworks, ESP, young learners including story-based teaching, BICS/CALP and Cummins, MTB-MLE, teen, exam prep, vocational).
+
06
Can ELT Brain replace ELT editors?

It speeds up the editorial process, provides consistency and brings SLA rigour at every stage. It runs the consistent rubric pass that senior editors do by hand, in seconds, across every page. But it doesn't remove the judgment work editors do – what to push back on, negotiating revisions with freelancers, calibrating across the unit and the series, and making the calls only humans can make.

The realistic outcome: every page gets the audit pass; editors apply human judgment to what the audit surfaces; freelancers get faster, more consistent feedback; the team scales without quality drift.

+
07
What if my workplace blocks custom AI connectors?

This is common at publishers and large institutions. Three paths usually still work:

1. Claude Code or the Code tab inside Claude Desktop – separate policy surface from Claude chat; plugin install is usually allowed.

2. Codex if you have a paid ChatGPT account – separate OpenAI app with its own settings.

3. Claude Desktop via local config – edits a local file on your machine and bypasses the "Custom connectors" setting entirely.

See the set-up guide for all three.

+
08
Will it work for specific ELT contexts?

The knowledge base covers SLA fundamentals plus context-specific notes for CLIL, EMI, EAL, ESOL, business English, BELF, EAP, ESP, young learners, teens, exam preparation, and vocational. The tools adapt to the CEFR level, L1, and instructional setting you specify.

If you work in a context that isn't covered – or is covered only partially – the tools say so rather than bluffing. We might be able to add it to the knowledge base. Get in touch: info@learningbrain.ai.

+
09
What if it pushes back on something I know works?

Good – that's a conversation worth having. The pushback is always grounded in a specific research finding you can evaluate against your contextual knowledge. Sometimes the research wins. Sometimes your practitioner judgment wins. Either way, the decision is now explicit rather than assumed.

+
10
Will it replace my editorial or design judgment?

No. The tools handle the rubric pass: level coherence, construct validity, lexical coverage, naturalness, inclusion. You handle the context: knowing your audience, calibrating your brand voice, negotiating with a freelancer, deciding what matters most for this series. No tool can guarantee a coursebook page will work in the classroom. It can guarantee the design is structurally defensible. What happens in delivery is still up to humans.

+
11
How is this different from just asking ChatGPT about ELT?

When you ask ChatGPT a question about ELT or SLA, it draws on its general training data – which includes outdated findings, popular myths (younger-always-better, native-speakerism as a default), and confident claims without evidence ratings. ELT Brain draws on a curated, evidence-rated knowledge base and validated rubrics. It also refuses to fall back to general AI knowledge when the substrate is silent on something. Honestly, if you try the same prompts in Claude or Codex with ELT Brain vs. the same prompts in vanilla ChatGPT you'll immediately see the difference.

+
§ 08 GET ACCESS

Start using ELT Brain.

Enter your email and you'll get access immediately. Connect to your AI tool and start designing or auditing in three minutes.

Sign up for access
Instant free access. Two minutes to set up.
Taking you to your setup guide…

Claude Desktop · Settings
01 Add the ELT Brain connector
SettingsCustomizeConnectorsAdd custom connector
🔗 https://eltbrain.com/mcp
02 Install the Skill file
SettingsCustomizeSkillsUpload a skill
📄 learningbrain-skill.md
ready – your AI now has the ELT specialist team