mondoTWIN Researcher
SHRARE Technology AG v0.7.4
Trial registries and guidance
From the developer
Declared capabilities
- Search registered clinical trials by condition, drug, sponsor, phase and status
- Read one trial's record: endpoints, arms, sponsors, outcome, registry records
- Count and compare trials with read-only SQL over the twin's tables
Declared limitations
- Unrelated request
- Patient records, which mondoTWIN neither holds nor accepts
- A write action the read-only plugin cannot perform
Not specified in the available declarations: pricing & account requirements.
Onboarding example
Find completed phase 2 and 3 trials of colchicine after myocardial infarction: enrolment, primary endpoint, stop reason.
Sources & dates
Developer declarations; not independently tested. Listing last checked Nov 18, 2026. Package observed Nov 18, 2026.
Description
From the marketplace listing
Seven skills for clinical researchers who design trials: benchmark similar registered trials, count and compare trials across the registries (phases, enrolment, endpoints, stop reasons, sponsors), report which primary endpoints the trials of an indication used and how often each succeeded, by phase, and which comparators they used, review a protocol section against comparable trials and ICH guidance, draft eligibility criteria from comparable trials, choose endpoints and estimands, and quote ICH or FDA guidance. Every trial and passage is cited with its registry id or guideline section. After installing, connect the bundled mondoTWIN server once: the assistant opens the sign-in page, where you sign in or register a mondoTWIN researcher account (a new account waits for approval).
Language: English · Automatically detected from descriptions.
Changes
No changes recorded yet. Follow this plugin to keep track of future changes.
First seen Nov 18, 2026. Last seen Nov 18, 2026.
Similar plugins
Similar plugins could not be prepared right now. Please try again in a moment.
Browse plugins →Keywords
Files & skills
File archives
Skill instructions
eligibility-criteria8.48 KB
---
name: eligibility-criteria
description: Draft inclusion and exclusion criteria for a clinical trial from the criteria of comparable registered trials, and propose concomitant-medication exclusions from what those trials prohibit and from the investigational drug's likely interactions with background and commonly co-prescribed drugs. Use when the user asks to write, draft, compare, tighten or broaden eligibility criteria, inclusion/exclusion criteria, or prohibited and restricted medications for a study.
---
# Eligibility criteria
Draft inclusion and exclusion criteria grounded in what comparable trials used, and ground the medication exclusions in the prohibited medications of those trials.
## Ground rules
1. Every trial, id and threshold in the answer comes from a mondoTWIN tool result. Never invent a trial or a threshold. mondoTWIN serves no drug-interaction screen: an interaction you propose from pharmacology is your reasoning, labelled "pharmacology reasoning, not tool-screened", never presented as a database finding.
2. Cite each trial by its registry id linked to its registry page: `https://clinicaltrials.gov/study/<NCT id>` for an NCT id, `https://www.isrctn.com/<ISRCTN id>` for ISRCTN, `https://www.clinicaltrialsregister.eu/ctr-search/search?query=<EudraCT number>` for EU CTR; other registries' ids unlinked, with the registry's name.
3. Say plainly when a search returned nothing, with the query used.
4. State the data as-of date from the tools, never from memory: each registry's `through_on` in `coverage` (`SELECT source, through_on FROM coverage WHERE entity_type = 'trial record'`) and the release id.
5. No patient data. Criteria are written for a population, never assessed for a named or identifiable patient. If the user pastes patient information, do not repeat it and ask them to remove it.
6. The output is a draft for qualified professionals (investigators, clinical pharmacologists, regulatory) to validate. Say so in one line at the end.
7. Answer in the user's language. Quote registry criteria verbatim in their original language.
8. Use only `search_clinical_studies`, `get_clinical_study`, `twin_sql` and `twin_record` (for the criteria text), `twin_text`, `search_ich_guidelines` and, for enrichment questions only, `get_disease_targets`. `twin_sql` also counts and compares across many trials (read-only SQL over the twin's tables; see the query-database skill). If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval.
## How the tools behave
**Trials.** `search_text` keeps the trials in which every word (or "quoted phrase") appears, each in any of the title, a condition, a drug or a sponsor as the registry wrote them: use the disease ("pulmonary fibrosis" in quotes keeps the phrase whole) or one to three distinctive words, plus `phases`, `statuses` and `icd11_chapters` filters (the chapter is the first character of a relevant hit's `icd11_code`: `B…` is "11", `C…` is "12"; it removes passing mentions). If a query returns 0, drop a word and retry before concluding anything. A phase filter drops trials that state no phase (`NA`). Read at most 6 trials with `get_clinical_study`. It returns no eligibility criteria: v1 has no criteria column.
**Criteria text.** The criteria are in the trial's `population` passage (the title, description and eligibility the twin's population model read):
`twin_sql(query="SELECT p.passage_id, left(p.text, 2000) AS text FROM document_link l JOIN Preview of the first 4,000 characters. Read full instructions →
endpoint-and-estimand8.79 KB
--- name: endpoint-and-estimand description: Summarise which primary and key secondary endpoints comparable registered trials used and at which time points, and draft the estimand for a trial objective following ICH E9(R1) (treatment, population, variable, intercurrent events with their strategies, population-level summary). Use when the user is choosing or justifying a primary or secondary endpoint, asks what similar trials measured and when, or asks to define, check or write an estimand or an intercurrent-event strategy. --- # Endpoints and estimand Show what comparable trials measured and when, then draft the estimand that the chosen endpoint serves, in the structure of ICH E9(R1). ## Ground rules 1. Every trial, id, endpoint, time point and guideline passage in the answer comes from a mondoTWIN tool result. Never invent a trial, a number or a quotation. Statistical reasoning is yours; label it as reasoning, not as evidence. 2. Cite each trial by its registry id linked to its registry page: `https://clinicaltrials.gov/study/<NCT id>` for an NCT id, `https://www.isrctn.com/<ISRCTN id>` for ISRCTN, `https://www.clinicaltrialsregister.eu/ctr-search/search?query=<EudraCT number>` for EU CTR; other registries' ids unlinked, with the registry's name. Cite each guideline passage by guideline, section (when visible) and a short verbatim quote. 3. Say plainly when a search returned nothing, with the query used. 4. State the data as-of date from the tools, never from memory: each registry's `through_on` in `coverage` (`SELECT source, through_on FROM coverage WHERE entity_type = 'trial record'`, through `twin_sql`) and the release id; ICH guidance as held in the mondoTWIN database. 5. No patient data. If the user pastes identifiable patient information, do not repeat it and ask them to remove it. 6. The output is a draft for qualified professionals (clinicians and statisticians) to validate. Say so in one line at the end. 7. Answer in the user's language. Quote registry text and guidelines verbatim in their original language. 8. Use only `search_clinical_studies`, `get_clinical_study`, `search_ich_guidelines`, `get_ich_guideline_full` and, for biomarker endpoints only, `get_disease_targets`. `twin_sql` is also available for counts and comparisons across many trials (read-only SQL over the twin's tables, for example how often an endpoint is primary; see the query-database skill), and for the criteria and arms text of a trial. If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval. ## How the tools behave **Trials.** `search_text` keeps the trials in which every word (or "quoted phrase") appears, each in any of the title, a condition, a drug or a sponsor as the registry wrote them: use one to three distinctive words (the drug, the disease), plus `phases`, `statuses` and `icd11_chapters` filters (the chapter is the first character of a relevant hit's `icd11_code`: `B…` is "11", `C…` is "12"; it removes passing mentions). If a query returns 0, drop a word and retry before concluding anything. A phase filter drops trials that state no phase (`NA`): run one search without it. Read at most 6 trials with `get_clinical_study`. Its `endpoints[]` have `kind` (primary or secondary), `position`, `measure`, `description` and `timepoint` (the time frame as the registry states it; NULL when it states none, then look in the measure or description). When no time point is given, write "time frame not stated". EU
Preview of the first 4,000 characters. Read full instructions →
endpoint-comparator-precedent11.2 KB
---
name: endpoint-comparator-precedent
description: Report which primary endpoints the registered trials of an indication used and how often each succeeded or failed, by phase, and which control arms and active comparators those trials used and how they fared, from mondoTWIN's twin. Use when the user asks which endpoints work or have succeeded for a disease, for endpoint precedent or success rates by endpoint, what primary endpoint a phase 3 should use, which comparator or control arm trials in the indication used, the standard-of-care comparator, placebo versus active comparator, or what trials in an indication have measured. Descriptive precedent, not a prediction.
---
# Endpoint and comparator precedent
Two design-time questions have an answer in the registered trials of an indication: which primary endpoints have held up, and which control arms analogous trials used. Answer both from the twin's tables, by phase, as counts a researcher can check row by row. The endpoint half needs a clustering step, because registries hold endpoint labels as free text; the comparator half is structured (`trial_drug.role`).
## Ground rules
1. Every trial, id, count and endpoint label in the answer comes from a mondoTWIN tool result. Never invent a trial or a number. The clustering of endpoint labels (step 2b) is your reading of the labels the tools returned: say so, and give the raw count behind each cluster.
2. Cite the release id (`SELECT release FROM release_info`) and, for any list of trials (the comparators, the trials of a small cell), their registry ids: `trial_record.record_key` where `is_primary`. Link an NCT id as `https://clinicaltrials.gov/study/<NCT id>`, an ISRCTN id as `https://www.isrctn.com/<ISRCTN id>`, a EudraCT number as `https://www.clinicaltrialsregister.eu/ctr-search/search?query=<EudraCT number>`; other registries' ids unlinked, with the registry's name.
3. Say plainly when a query returned nothing, with the query used.
4. State the data as-of date from the tools, never from memory: each registry's `through_on` in `coverage` (`SELECT source, through_on FROM coverage WHERE entity_type = 'trial record'`).
5. No patient data. If the user pastes identifiable patient information, do not repeat it and ask them to remove it.
6. The output is descriptive precedent for qualified professionals (clinicians, statisticians, regulatory) to weigh; it is not a prediction and no success-probability model is applied. Say so in one line at the end.
7. Answer in the user's language. Quote endpoint labels as the registries wrote them.
8. Use only `twin_find`, `twin_catalog`, `twin_sql`, `twin_text` and, to read one trial, `get_clinical_study`. `twin_sql` runs one read-only `SELECT` or `WITH` over the twin's tables (15 s, at most 500 rows; see the query-database skill); page with `OFFSET` when a reply says `more`. `twin_catalog(table="trial_endpoint")` shows a table's columns. If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval.
## Step 1: resolve the indication to ICD-11 codes
The twin keys conditions on ICD-11 (`icd11:<code>`) and Orphanet (`orpha:<code>`). Resolve the name first, and show the candidates: a name often spans several codes, and not all are the same indication. The trials' own wording shows which codes they were coded to:
```
twin_sql(query="
SELECT c.icd11_id, max(d.name) AS code_title, max(c.raw_text) AS example_text,
count(DISTINCT c.trial_id) AS trials
FROM triPreview of the first 4,000 characters. Read full instructions →
guidance-lookup5.59 KB
---
name: guidance-lookup
description: Answer a question about ICH or FDA regulatory guidance with the exact passage quoted, the guideline id and the section. Use when the user asks what ICH (for example E6(R3), E8(R1), E9(R1), E10, M11, the Q series) or FDA guidance says about a topic, wants the source or exact wording of a regulatory expectation, or wants to check a statement against the guidance.
---
# Guidance lookup
Find the passage in ICH or FDA guidance that answers the question, quote it, and say exactly where it comes from.
## Ground rules
1. Every quotation, guideline id and section in the answer comes from a mondoTWIN tool result. Never invent or paraphrase a passage as if it were a quote. You may name a document you believe is relevant but did not retrieve; label it "not retrieved from the database".
2. Cite each passage by guideline (with its revision), section (when visible) and a verbatim quote. Add the passage number (`chunk_id`) so the user can find it again.
3. Say plainly when the search found nothing relevant, with the queries used. The corpus is not complete.
4. State that the answer uses ICH and FDA guidance as held in the mondoTWIN database, and that the user should confirm the current version on ich.org or fda.gov.
5. No patient data. If the user pastes identifiable patient information, do not repeat it and ask them to remove it.
6. The output is a draft for qualified professionals (regulatory affairs) to validate. Say so in one line at the end.
7. Answer in the user's language. Quote the guidance verbatim in English; add a translation when the user's language differs.
8. Use only `search_ich_guidelines`, `get_ich_guideline_full`, `search_fda_guidelines` and `get_fda_guideline_full`. `twin_sql` is also available for counts and comparisons across many trials (read-only SQL over the twin's tables; see the query-database skill). If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval.
## How the tools behave
**ICH.** `search_ich_guidelines` is a semantic search that returns passages with a bare `guide_type` and a `chunk_id`. Known mappings: `E6` holds E6(R3) (it uses "participant" and numbers such as 2.8.5); `E8` holds E8(R1); `E9` holds E9(R1), with the estimands addendum in the A sections (for example A.3.2 strategies, A.5.2 sensitivity analysis); `M11` holds the Step 2 draft guideline and template (2022, template draft updated March 2025): cite it as "M11 (Step 2 draft)". For other codes, take the revision from the passage text or say it is not visible. There is no section field: read the section number from a heading inside the passage ("## A.3.2", "3.3", "2.10.3"). If none is visible, search again with the passage's heading words or opening sentence, or write "section not identified".
**FDA.** `search_fda_guidelines` returns passages with a `filename`, which is the document title with underscores (write it as a title). Draft or final status shows only when the passage contains "Draft — Not for Implementation" or a date. The corpus lacks some guidance documents (for example the guidance on adaptive designs for drugs and biologics). A passage that only mentions or cites another guidance is not that guidance: say the cited document is not in the database rather than answering from the citing text.
**Queries.** Phrase queries in the guideline's own vocabulary (for example "interim analysis", "group sequential", "estimand", "intercurrent event", "non-inferiority mPreview of the first 4,000 characters. Read full instructions →
protocol-section-review8.8 KB
--- name: protocol-section-review description: Review a draft clinical trial protocol or synopsis section (objectives, endpoints, eligibility criteria, sample size rationale, schedule of activities) against comparable registered trials and ICH E6(R3), E8(R1), E9(R1) and M11, and return numbered, reasoned changes, each backed by trial registry ids or a quoted guideline passage and marked "to validate". Use when the user pastes or describes part of a protocol and asks for a review, critique, gap check, improvements, or consistency with ICH or with what similar trials did. --- # Protocol section review Check one protocol section against what comparable trials did and what ICH guidance says, and propose concrete, evidenced changes. ## Ground rules 1. Every trial, id, number, date and guideline passage in the answer comes from a mondoTWIN tool result. Never invent a trial, a value or a quotation. Background knowledge may shape the review and the search terms, but a statement presented as evidence must come from a tool result. 2. Cite each trial by its registry id linked to its registry page: `https://clinicaltrials.gov/study/<NCT id>` for an NCT id, `https://www.isrctn.com/<ISRCTN id>` for ISRCTN, `https://www.clinicaltrialsregister.eu/ctr-search/search?query=<EudraCT number>` for EU CTR; other registries' ids unlinked, with the registry's name. Cite each guideline passage by guideline, section (when visible) and a short verbatim quote. 3. Say plainly when a search returned nothing, with the query used. 4. State the data as-of date from the tools, never from memory: each registry's `through_on` in `coverage` (`SELECT source, through_on FROM coverage WHERE entity_type = 'trial record'`, through `twin_sql`) and the release id; ICH and FDA guidance as held in the mondoTWIN database. 5. No patient data. A protocol is fine; identifiable patient information is not. If the user pastes it, do not repeat it and ask them to remove it. 6. The output is a draft for qualified professionals to validate. Every proposed change is marked "to validate". 7. Answer in the user's language. Quote registry text and guidelines verbatim in their original language; add a translation when the user's language differs. 8. Use only `search_clinical_studies`, `get_clinical_study`, `search_ich_guidelines`, `get_ich_guideline_full`, `search_fda_guidelines`, `get_fda_guideline_full` and, for a mechanistic rationale only, `get_disease_targets`. `twin_sql` and `twin_record` read a trial's criteria and arms text, and `twin_sql` also counts and compares across many trials (read-only SQL over the twin's tables; see the query-database skill). If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval. ## How the tools behave **Trials.** `search_text` keeps the trials in which every word (or "quoted phrase") appears, each in any of the title, a condition, a drug or a sponsor as the registry wrote them, so use one to three distinctive words (drug, disease), plus `phases`, `statuses` and `icd11_chapters` filters (the chapter is the first character of a relevant hit's `icd11_code`: `B…` is "11", `C…` is "12"; it removes passing mentions). If a query returns 0, drop a word and retry before concluding anything. Trials that state no phase (`NA`) are dropped by a phase filter: run one search without it. `enrollment_actual` is the stated final count and `enrollment_planned` the target; a trial carries one or the other. `outcome`, `clinical_ou
Preview of the first 4,000 characters. Read full instructions →
query-database23.6 KB
---
name: query-database
description: Use when a question is to be answered from MondoTwin's twin of the pharma world through the twin MCP tools (twin_catalog, twin_find, twin_sql, twin_text, twin_record): a count, list, landscape or timeline of trials, drugs, products, approvals, indications, exclusivity, generics, companies, ownership, filings, papers, diseases or prevalence; reading a raw registry record, a label's text or an abstract; and before any analysis that cites the twin.
when_to_use: Trigger phrases include "find trials for", "how many trials", "drugs tested for", "trials of <drug>", "approved where", "authorised in", "indications of", "what the label says", "loss of exclusivity", "generics", "who holds", "who owns", "the sponsor's trials", "the group's pipeline", "what did the trial find", "why was it stopped", "papers reporting", "prevalence", "ICD-11 code", "Orphanet", "as of <date>", "twin_sql", "twin_find", "query the twin", "query the database". The frontmatter fields stay on one line each.
---
# Query the database
## What you are querying
MondoTwin's twin of the pharma world: 58 tables (schema v1), served read only as PostgreSQL views, one active release at
a time. `release_info` names the release and the capture date of every source (`inputs.captured`); `catalog` describes
every table (`about`) and every column; `coverage` says which source was read, for which scope, through which date.
Table names are bare on the search path: write `trial`, never a schema prefix.
Three facts shape every answer.
- **Compiled.** A value is a decision (a code chosen, a drug linked, an outcome classed) with a measured accuracy per
method. A `confidence` column is that measurement, never a per-row probability.
- **Dated on two axes.** What the source states (`event_on`, `start_date`, `approved_on`) and when the twin knew it
(`known_from`). Only the tables whose `about` says so support `known_from <= <date>` as an as-of filter
(`trial_outcome`, `regulatory_event`); `trial` holds today's state and its `known_from` is the first posting.
- **Partial on purpose.** Thirteen tables are published empty until a source is read: `deal`, `deal_party`, `payment`,
`right_grant`, `royalty`, `royalty_party`, `royalty_paid`, `financial`, `org_runway`, `price`, `spend`, `claim`,
`opportunity`. An empty result from them is "unknown", never "none".
Every row carries `sources` (record keys `<source>:<id>`, readable with `twin_record`) and `known_from`. Cite both,
with the release id, for every fact you take.
## The five tools
| Tool | What it does | Limits |
|---|---|---|
| `twin_catalog(table)` | the release ids and the tables whose name contains the word, with `about`, columns and comments | `table=""` returns about 80 KB and omits the columns of 36 tables: ask per word (`"trial"` gives the nine trial tables). The same text is in `catalog`: `SELECT about FROM catalog WHERE table_name = '...'`; for the columns use `SELECT c->>'name', c->>'comment' FROM catalog, jsonb_array_elements(columns) c WHERE table_name = '...'` (the `columns` cell itself is cut at 2,000 characters) |
| `twin_find(text, limit=20)` | a name or an outside id to the twin's ids: exact case-insensitive matches first (`via: name` or `external_id`), then word-prefix matches | up to 100. Outside ids as `scheme:value`: `nct:NCT…`, `pmid:…`, `doi:…`, `unii:…`, `chembl:…`, `cas:…`, `rxcui:…`, `drugbank:DB…` (11,168 drugs carry one), `cik:0000310158`, `lei:…`, `icd11:DD71`, `orpha:771`. A brand returns the drug and one `product` per territory; an INN returns the moiety and its salts and forms; a common word returns trials whose title starts with it: take the hit of the entity type you want |
| `twin_sql(query, max_rows=100)` | one read-only `SELECT`, `WITH`, `VALUES` or `TABLE` over the views | 15 s statement timeout; `max_rows` 1 to 500; the reply stops near 60 KB (`"more": true`, a `note`) and cuts a cell over 2,000 characters; one query at a time. `{"error": ...}` iPreview of the first 4,000 characters. Read full instructions →
Referenced files: 2 — file references are metadata, not package contents.
similar-trials-benchmark9.92 KB
--- name: similar-trials-benchmark description: Benchmark a planned or existing clinical trial against comparable registered trials (ClinicalTrials.gov, EU CTR/CTIS, ISRCTN, ANZCTR, UMIN) and return a table with registry id and link, phase, status, enrolment, primary endpoint and time frame, key eligibility criteria and stop reason. Use when the user asks for similar, comparable, competing or precedent trials, a trial landscape for a drug, indication or design, or "what has been done before" for a study they are planning. --- # Similar-trials benchmark Find the registered trials most comparable to the user's study and present them as one evidence table that a clinical researcher can check row by row. ## Ground rules 1. Every trial, id, number and date in the answer comes from a mondoTWIN tool result. Never invent a trial or a value. Background knowledge may suggest search terms (a drug class, a trial acronym), but a trial appears in the answer only if a tool returned it. When a field is empty, write "not stated in the record". 2. Cite each trial by its registry id linked to its registry page: `https://clinicaltrials.gov/study/<NCT id>` for an NCT id, `https://www.isrctn.com/<ISRCTN id>` for ISRCTN, `https://www.clinicaltrialsregister.eu/ctr-search/search?query=<EudraCT number>` for EU CTR. For other registries give the id and registry name unlinked. 3. Say plainly when a search returned nothing, with the query and filters used. "Not in the database" is not "does not exist". 4. State the data as-of date from the tools, never from memory: each registry's `through_on` in `coverage` (one `twin_sql` call: `SELECT source, through_on FROM coverage WHERE entity_type = 'trial record'`) and the `release` that `get_clinical_study` returns. Later registrations and updates are missing. 5. No patient data. Work at protocol and registry level only. If the user pastes identifiable patient information, do not repeat it and ask them to remove it. 6. The output is a draft for qualified professionals to validate. Say so in one line at the end. 7. Answer in the user's language. Keep registry ids, trial titles and quoted registry text in their original language. 8. Use `search_clinical_studies` and `get_clinical_study` to find and read trials, and `twin_sql` or `twin_record` for the eligibility text they do not return. `twin_sql` also counts and compares across many trials (read-only SQL over the twin's tables; see the query-database skill). If the mondoTWIN tools are not available, or a call answers that authorization is required, the mondoTWIN connector is not connected yet: never answer from memory as though from the database. Tell the user in one short message how to connect it, and continue once the tools answer: in Claude, select Connect when Claude offers it, or Customize > Connectors > mondotwin > Connect; in ChatGPT, Plugins > mondoTWIN Researcher (or the mondoTWIN connector) > Connect, then start a new chat. They allow access, then sign in or register with a mondoTWIN account; a new account waits for mondoTWIN's approval. Do not answer from memory as though from the database. ## How the trial tools behave - `search_text` keeps the trials in which every word (or "quoted phrase") appears, each in any of the title, the brief title, a condition, a drug or a sponsor as the registry wrote them, or matches a registry id. "colchicine stroke" keeps trials that mention both words; `"pulmonary fibrosis"` in quotes keeps the phrase whole. Start with one to three distinctive words (the drug, the disease) plus filters. A word the registry spells differently (an abbreviation, "MI" for myocardial infarction) removes good matches: if a query returns 0 or too few, drop a word and retry before concluding anything. - Results come newest first (by start date); `total` is the match count; use `page_size` 10 to 20. Many matches name the word in passing (background therapy, a co-condition): screen by title, `diseases` and `drugs`. - Filters: `phases` (a ph
Preview of the first 4,000 characters. Read full instructions →
Release notes
0.7.4: all seven skills tell the user exactly where to connect the mondoTWIN server when its tools are missing (Claude: the Connect prompt or Customize > Connectors; ChatGPT: Plugins > mondoTWIN Researcher > Connect, then a new chat), instead of stopping. 0.7.3: the skills ask the user to connect the mondoTWIN server when its tools are missing, instead of stopping; the listing says the connector is connected once after installing. 0.7.2: listing category Education & Research and seven capability lines for the ChatGPT directory; no change to the skills. 0.7.1: query-database repaired: 0.7.0 shipped a damaged SKILL.md (method steps carrying rule text, the question-to-table router and the rules section missing); the skill is whole again, seven sections, rules R1 to R18, every recipe run on the served release. 0.7.0: query-database rewritten: the five twin tools with their measured limits, a seven-step method, a question-to-table router, 18 rules from the live catalog and 24 recipes each run on the served release, with the 58 tables one line each; nothing but the twin. 0.6.0: the connector moves to https://researchers.mondotwin.com/mcp, the researchers door: the same twelve read-only tools to every account; reconnect the mondoTWIN connector once. 0.5.0: endpoint-comparator-precedent, a new skill: which primary endpoints the registered trials of an indication used and how often each succeeded or failed, by phase, and which control arms and active comparators they used, from the twin's tables, with the n behind every rate. 0.4.0: query-database replaced query-clinical-studies: one read-only skill over the twin's tables (trials, drugs and products, approvals and indications, exclusivity, companies and filings, papers, diseases), its recipes run against the served release, the same text the platform's own analysts use. Seven skills for trial design research on mondoTWIN's twin of registered trials (ClinicalTrials.gov, EU CTR/CTIS, ISRCTN, ANZCTR, UMIN, ICTRP) and ICH and FDA guidance, with the mondoTWIN connector.
Declared in the saved package. Remote tools may change independently.
Package details
Publisher declarations from the archived package. These are separate from our research and the live service's terms.
- Package author
- SHRARE Technology AG / mondoTWIN
- Keywords
- See keywords
- Commerce declaration
- Does not support commerceThis does not establish whether access is free or paid.
- Review scenarios
- 5 positive · 3 negativeScenario counts, not test results.
Declared capabilities
- Search registered clinical trials by condition, drug, sponsor, phase and status
- Read one trial's record: endpoints, arms, sponsors, outcome, registry records
- Count and compare trials with read-only SQL over the twin's tables
- Endpoint and comparator precedent for an indication, by phase
- Draft eligibility criteria from comparable trials
- Quote ICH and FDA guidance with the guideline and section
- Genetic targets of a disease from Open Targets
Package observed Nov 18, 2026.
Technical details
- First seen
- Nov 18, 2026 · 06:00 UTC
- Last seen
- Nov 18, 2026 · 06:00 UTC
- Collection status
- Collected
plugin_asdk_app_6aea91e28adc8191aaf959530f596136
Download plugin data (JSON)Plugin details: freshly prepared. Prepared . Refreshed when plugin data changes or after 6 hours. Your account and Follow status are live.