ELI5 — connector documentation

• MCP endpoint: https://eli5.connectorbay.com/mcp
• REST API: POST https://eli5.connectorbay.com/v1/tools/{tool_name}, described by the OpenAPI 3.1 spec at https://eli5.connectorbay.com/openapi.json
• Transport: Streamable HTTP, stateless, MCP specification 2025-06-18 or later
• Authentication: none required. An optional Authorization: Bearer <key> raises rate limits.
• Publisher: Unify Dynamics LLC · connectorbay.com/eli5 · agent@connectorbay.com

What this connector is for

People routinely hold documents they cannot read: an Explanation of Benefits, a lease, a loan disclosure, a lab result, a benefits notice, thirty pages of terms of service. They also need things explained — to themselves, and to their children.

An agent can already write a good explanation. What it cannot do on its own is look up a current authoritative definition and cite it, compute a reading grade from published formulas, notice that clause 14(b) is a class-action waiver, or check whether the four numbers on a medical bill actually add up. ELI5 does those things and hands the results to the agent, which does the writing.

That division is deliberate, and it is also why every tool here is safe: the connector reads, computes, and returns. It sends nothing, pays nothing, files nothing and changes nothing.

Tool reference

Decode

decode_document(text, document_type?)

The general entry point for "what does this mean". Detects which of fifteen document types it is (Explanation of Benefits, medical bill, lab report, prescription, lease, loan, credit card statement, terms of service, privacy policy, insurance policy, employment agreement, government notice, utility bill, tax document, other), then returns:

• the glossary terms present, each with a plain definition and why it matters
• jargon with no definition, and acronyms used without being spelled out
• what the document asks of you, the deadlines, and the dates
• the clause flags, by severity
• readability: consensus grade level, reading ease, and the hardest sentences with reasons
• what to watch for and what to ask, specific to the document type

decode_money_document(text, document_type?)

Everything above, plus arithmetic. Labelled dollar amounts are bucketed into billed, adjustment, allowed, plan-paid, deductible, copay, coinsurance and patient responsibility, and then checked:

decode_medical_text(text, look_up_terms?)

Parses result lines of the form Name value unit (low–high) and flags each as within, above or below its printed range, with how far outside as a fraction of the range's own width. Dates and times are masked first so 2026-09-02 never parses as a result. Terminology is looked up in MedlinePlus, which NLM writes for patients rather than clinicians.

It explains words. It does not interpret results, suggest a diagnosis, or advise on medication, and its response instructs the agent accordingly.

find_the_catch(text)

Scans for 25 clause patterns, grouped by severity:

High — binding arbitration · class action waiver · jury trial waiver · automatic renewal · unilateral change of terms · prepayment penalty · balloon payment · deficiency balance · personal guarantee · confession of judgment · one-way indemnification

Medium — limitation of liability · broad content licence · data sharing or sale · early termination fee · late fees and default interest · acceleration · attorney-fee shifting · distant venue or foreign law · non-compete · invention assignment · non-refundable deposit

Informational — landlord entry rights · forced updates or feature removal · survival clauses

Each match returns the language found, what the clause means in practice, and the question to ask.

explain_term(term, domain?)

Checks the bundled glossary (83 terms across insurance, lending, credit, legal, housing, employment, tax, investing, benefits, medical, government, technology and consumer), then MedlinePlus for medical terms, then Wiktionary and Simple English Wikipedia.

Explain

explain(topic, level) and explain_levels(topic, levels?)

Levels are age5, age10, teen, adult, expert. Both return grounded summaries, the vocabulary that will block the reader, analogy scaffolds, common confusions and a concrete writing target — sentence length, vocabulary ceiling, target grade band.

explain_levels is the signature call: one topic prepared for five audiences on shared grounding. The rungs are meant to differ in what they contain, not merely in vocabulary — the age-5 version leaves the mechanism out, the expert version leads with the edge cases.

analogy_for(concept, audience)

Matches the concept to a structural shape — a rate, compounding, a threshold, a network, a feedback loop, selection, risk pooling, layered abstraction, a trade-off frontier, signal and noise, stock versus flow, map versus territory, authentication, queueing, immune memory — and returns vehicles that share that shape.

Every vehicle carries a breaks field. This is the point of the tool. An analogy handed over without its limits is how misconceptions are manufactured: tell someone exponential growth is like folding paper and they will conclude that exponentials stop on their own, because paper folding does.

readability_score(text)

Flesch Reading Ease, Flesch-Kincaid Grade, Gunning Fog, SMOG, Coleman-Liau, Automated Readability Index, New Dale-Chall and LIX, plus a consensus grade (the median of the grade formulas defined for that text), the hardest sentences with the reason each is hard, and the unfamiliar words.

Syllables come from the CMU Pronouncing Dictionary rather than from hyphenation, which is why these numbers differ from many online checkers — hyphenation-based counters undercount badly (they give "deductible" two syllables and "idea" one). Validation: the Gettysburg Address scores 65.9 here against its published Flesch value of about 66.

SMOG is omitted below 30 sentences and the result is marked low-confidence below 50 words, because every readability formula is noise at that length.

simplify_plan(text, target_grade)

An ordered plan rather than a rewrite: sentences to split, word swaps marked safe or meaning-changing, passive constructions, buried verbs, padding to cut, undefined acronyms — in the order that moves the score most. Sentence length first, because it moves every formula at once.

define_simply(word, for_level) and jargon_scan(text, limit)

define_simply filters synonyms by actual corpus frequency, so the "simpler" word offered is one the reader plausibly knows. jargon_scan ranks jargon by a blocking score weighting frequency in the text, word length, and whether we can define it.

A word counts as jargon only if it is outside both the Dale-Chall familiar list and the 6,000 commonest modern English words. Dale-Chall dates from the 1940s and flags ordinary modern words like "information" and "submit"; scores use it unmodified for fidelity to the published formula, but the lists shown to a user are filtered, so the jargon that surfaces is jargon.

Learn

explain_for_grade(topic, grade) · grade_level_target(grade)

Grade bands follow the Common Core Appendix A stretch bands, with sentence-length and syllable targets from the readability literature. Both tools state the caveat plainly: bands are guidance, not gates, and a child reading about something they love reads two bands above them.

common_misconceptions(topic, grade?)

Thirty catalogued misconceptions across science and mathematics, drawn from the education research literature — AAAS Project 2061, the Force Concept Inventory tradition, and the Rational Number Project work on fractions and decimals. Each gives the wrong idea, the correction, why the wrong idea is appealing, and a diagnostic question.

Use the diagnostic question before teaching. An explanation laid over an unaddressed misconception usually bounces off.

check_understanding(topic, grade, levels?)

Question stems laddered by Bloom's taxonomy (Anderson & Krathwohl 2001), scaled to the grade, plus diagnostics for the known misconceptions. It also names the most common false positive: a learner repeating your phrasing back. Ask for a new example, not a restatement.

word_ladder(word, grade)

Plain definition, the simplest usable stand-in word, and the prefix, root and suffix with what each part means — including the Greek and Latin combining forms (photo-, thermo-, hydro-, bio-, neuro-, -scope, -graph) that unlock whole families of school vocabulary at once.

is_this_right_for_my_kid(text, grade)

A verdict, not a score: comfortably below level, about right, a stretch, or too hard unaided — with the specific sentences and words that make it so, and the words worth pre-teaching.

Access, limits and reliability

Free tier — no account or API key. 300 tool calls per user per day, 60 per minute, keyed on the user identifier supplied by the host agent where the host sends one, otherwise the client IP. Pro tier — optional API key, Authorization: Bearer <key>, 5,000 calls per day.

Exceeding a limit returns HTTP 429 with a Retry-After header and a structured JSON-RPC error. initialize and notifications are never counted.

Degradation. Upstreams are public, keyless and rate-limited, so outages are a normal condition rather than an exception. A tool that loses one optional source returns its result with a degraded list naming what was missing. A tool that loses all of its sources returns the bundled scaffolding with grounding_unavailable: true and an instruction to the agent to explain from its own knowledge, say so to the user, and cite nothing. Everything computed from bundled data — readability, the glossary, clause detection, grade bands, misconceptions, analogies — works with no network at all.

Geography. The upstreams are U.S. English-language sources; the connector is usable anywhere but its material is U.S.-centred, which the document types and glossary reflect.

Audience. General. Several tools are built for parents and teachers working with children; nothing here is age-gated and no tool collects information from a child.

Sources

Wikipedia and Simple English Wikipedia (CC BY-SA 4.0, attributed in every response) · Wiktionary (CC BY-SA) · Datamuse API · MedlinePlus, U.S. National Library of Medicine · openFDA, U.S. Food and Drug Administration · RxNorm, NLM · the Dale-Chall list of 3,000 familiar words · the CMU Pronouncing Dictionary · google-10000-english, from Google's n-gram frequency data (CC BY 3.0) · plain-language substitutions after the federal plain-language guidelines and the Plain Writing Act of 2010 · Common Core State Standards Appendix A text-complexity bands · Bloom's taxonomy, Anderson & Krathwohl 2001 revision.

Glossary definitions are restated in plain English from U.S. public-sector sources (CMS, HealthCare.gov, CFPB, SEC Investor.gov, IRS, DOL, HUD, FTC), not quoted from them.

Limits and disclaimers

General information only. Not legal, medical, tax or financial advice. Clause detection is pattern matching, not legal analysis: a match means language worth reading closely, a miss does not prove a clause is absent, and state law makes several of these clauses unenforceable regardless of what a document says. Medical tools explain terminology and do not interpret results or advise on treatment. Readability formulas measure sentence length and word familiarity; they cannot see whether ideas or subject matter suit a reader.