A Model Context Protocol server for Zotero integration. It gives any LLM full access to your Zotero library: search, organize, add papers by DOI, import PDFs, read full-text content, and inject live citations into Word documents. The server is designed to be usable by any LLM without external documentation. On connection, it sends workflow instructions via the MCP instructions field, and each tool
Add this skill
npx mdskills install Xevos117/scientific-writingComprehensive workflow with evidence-based writing, transparent source triage, and Zotero citation integration
1---2name: scientific-writing3description: >4 Use this skill whenever the user asks to write a scientific article, literature review,5 research paper, case report, systematic review, or any academic/scholarly document.6 Triggers include: "write a paper", "write an article", "literature review", "review article",7 "case report", "research paper", "scientific article", "academic paper", or requests to8 produce documents citing scientific literature. This skill enforces a structured research9 workflow with source triage, evidence-based writing, and transparent disclosure of which10 sources were read in full vs. abstract only. Also use when the user asks to summarize,11 synthesize, or compare multiple scientific papers into a written deliverable.12---1314# Scientific Article Writing Skill1516## Core Principle1718**Abstracts are valid sources. Be transparent about what you used.**1920Abstracts contain legitimate, citable information: quantitative results, study21design, sample sizes, effect sizes, p-values, and main conclusions. If the22abstract provides the data you need, there is no obligation to retrieve the23full text — doing so consumes context without adding value.2425However, **transparency is non-negotiable**. The user must always know which26sources were read in full and which were used via abstract only. Before27presenting the final document, include a clear disclosure listing abstract-only28sources (see Source Transparency below).2930Full-text reading remains valuable for interpretive claims, methodological31details, subgroup analyses, and limitations not covered in the abstract. Use32your judgment: if the abstract is sufficient, use it; if you need more depth,33attempt full-text retrieval.3435### PDF Upload Policy3637PDF upload to Zotero happens **only when the user explicitly requests it**:3839- **If the user explicitly requests PDF upload** → it is **mandatory and cannot40 be skipped**. Follow the verify-upload-validate procedure in the41 `zotero-mcp-integrations` skill. For paywalled sources, present a table to the42 user asking them to upload the PDFs manually if they have access.43- **If the user does NOT request PDF upload** → do not upload PDFs.4445### Source Transparency (MANDATORY)4647Before presenting the final document, you **MUST** disclose which sources were48used via abstract only. A simple list is sufficient:4950> **Sources used via abstract only:** Smith et al. 2021, Jones et al. 2023.51> All other sources were read in full.5253This is not a warning — it is an informational disclosure that lets the user54judge the evidence quality and decide whether to provide additional full texts.5556---5758## Workflow Overview5960The workflow has 5 mandatory phases executed in strict order.61No phase can be skipped. Each phase has explicit entry/exit criteria.6263```64Phase 1: SCOPE → Phase 2: SEARCH → Phase 3: READ → Phase 4: WRITE → Phase 5: DELIVER65```6667See `references/workflow-detail.md` for the complete phase specifications.6869---7071## Phase 1: SCOPE — Define the Article7273**Goal**: Understand exactly what needs to be written before searching anything.7475**Actions**:761. Identify article type (review, research, case report, systematic review, meta-analysis, editorial, letter, etc.)772. Determine target audience and field783. Determine approximate length and section structure794. Identify citation style (Vancouver, APA, IEEE, etc.)805. Clarify key questions the article must answer816. Agree on scope boundaries — what is IN and what is OUT8283**Exit criteria**: Clear written specification of what will be produced.8485**Anti-pattern**: Starting to search before knowing what you're looking for.86This leads to unfocused literature retrieval and wasted effort.8788---8990## Phase 2: SEARCH — Find Sources9192**Goal**: Build a comprehensive, relevant bibliography.9394**Actions**:951. Construct search queries based on the scope (use PubMed, web search, or whatever tools are available)962. Cast a wide initial net — search for more sources than you'll ultimately cite973. Triage results: categorize by relevance (essential / supporting / background)984. Record all sources with their identifiers (DOI, PMID, URL)99100**Search strategy by article type**:101102| Article Type | Minimum Sources | Search Depth |103|---|---|---|104| Narrative review | 15-30 | Broad, thematic |105| Systematic review | Exhaustive per protocol | Protocol-driven, reproducible |106| Research article | 10-25 | Focused on methods + context |107| Case report | 5-15 | Similar cases + guidelines |108| Short communication | 5-10 | Targeted, recent |109110**Exit criteria**: A ranked list of sources with abstracts (and full text where easily available).111112**Anti-pattern**: Stopping at 10 PubMed results.113114---115116## Phase 3: READ — Understand the Sources117118**Goal**: Develop sufficient understanding of the source material before writing.119120### When abstracts are enough121122For many sources — especially supporting or background references — the abstract123provides all the data you need: study design, sample size, key results, main124conclusion. In these cases, there is no need to retrieve the full text.125126### When full text adds value127128For **essential** sources where you need interpretive depth, methodological detail,129subgroup analyses, or limitations, attempt full-text retrieval:1301311. **web_fetch on DOI URL** — often resolves to full HTML text on publisher sites1322. **Open access repositories** — PMC, Europe PMC, BioRxiv, MedRxiv1333. **Publisher open access** — JMIR, PLOS, BMC, Frontiers, MDPI are fully OA1344. **Preprint versions** — BioRxiv, MedRxiv, arXiv, SSRN135136If full text is behind a paywall, the abstract is sufficient. Do not waste137effort trying to circumvent access restrictions.138139### What to extract from each source140141Whether from abstract or full text, for each source you should know:142143- What was the study question?144- How was it designed? (sample size, population, key methods)145- What were the main numerical findings?146- How does this fit into the broader evidence base?147148For full-text sources, also extract:149- Methodological details not in the abstract150- Stated limitations151- Subgroup analyses or secondary endpoints152153### Track your source depth154155Keep a mental note of which sources you read in full and which you used via156abstract only. You will need this for the Source Transparency disclosure in157Phase 5.158159**Exit criteria**: Sufficient understanding of each source to make accurate claims.160161**Anti-pattern**: Paraphrasing abstract conclusions without extracting specific data.162Even when using abstracts, cite concrete numbers, not vague summaries.163164---165166## Phase 4: WRITE — Compose the Article167168**Goal**: Produce a well-structured scientific article grounded in the sources you've read.169170### Structure by Article Type171172See `references/article-structures.md` for detailed templates.173174**General principles that apply to all types**:1751761. **Introduction**: Establish context → Identify the gap/need → State the objective1772. **Body**: Organize by theme, chronology, or methodology — NOT by source.178 Never write "Smith et al. found X. Jones et al. found Y. Lee et al. found Z."179 Instead: "Multiple studies have demonstrated X [refs], although the effect varies180 by population [ref1, ref2]."1813. **Discussion/Synthesis**: Integrate findings, identify patterns, acknowledge conflicts1824. **Conclusions**: Answer the question posed in the introduction183184### Writing Rules185186- **Every factual claim needs a citation**. No exceptions.187- **Use specific data**, not vague summaries. Say "92.5% maintained viral188 suppression at 48 weeks" not "most patients responded well." Abstracts189 often contain these specific numbers — use them.190- **Acknowledge contradictions** in the literature. Don't cherry-pick.191- **Distinguish between what the data shows and what it suggests**. Use language192 appropriately: "demonstrated" vs "suggested" vs "may indicate."193- **State limitations explicitly**, both of individual studies and of the review itself.194- **Track citation numbers** carefully if using numbered styles (Vancouver, IEEE).195 Each unique source gets one number, assigned in order of first appearance.196197### Citation Placement198199- Citations go AFTER the claim, BEFORE the period: "...maintained suppression [4]."200- Multiple citations: [4,5] or [4-6] for ranges201- Abstract-only sources can be cited for specific factual claims directly stated202 in the abstract (quantitative results, study design, sample size, main conclusion).203 Do NOT use abstract-only sources for interpretive or methodological claims that204 require full-text context.205206**Exit criteria**: Complete draft with all citations placed.207208**Anti-pattern**: Writing the full text and then "sprinkling" citations afterward.209Citations should be integral to the writing process, not a post-hoc decoration.210211---212213## Phase 5: DELIVER — Produce the Final Document214215**Goal**: Generate a polished, properly formatted deliverable.216217**Actions**:2181. Generate the document in the requested format (.docx, .pdf, .md, etc.)2192. If Zotero integration is available, inject citation field codes2203. **Include the Source Transparency disclosure** (see Core Principle) listing221 which sources were used via abstract only2224. If the user requested PDF upload, present a table of paywalled sources they223 may want to upload manually (see `zotero-mcp-integrations` skill)2245. Validate the document2256. Present to user with clear instructions for any post-processing (e.g., Zotero → Add/Edit Bibliography, then Zotero → Refresh)226227**Exit criteria**: Downloadable document with working citations.228229---230231## Tool-Agnostic Principles232233This skill works regardless of which tools are available. The workflow is the same whether you have:234235- PubMed MCP tools + Zotero MCP tools + web_search236- Only web_search + web_fetch237- Only web_search238- No search tools at all (user provides sources)239240The key insight is: **the workflow is about the PROCESS, not the tools.**241242| If you have... | Use it for... |243|---|---|244| `pubmed_search_articles` | Phase 2: Structured literature search |245| `pubmed_fetch_contents` | Phase 2-3: Get abstracts and metadata |246| `web_search` | Phase 2: Find sources; Phase 3: Locate full text when needed |247| `web_fetch` | Phase 3: Read full-text HTML when abstract is not sufficient |248| `import_pdf_to_zotero` | Phase 5: Archive PDFs when user requests it (see zotero-mcp-integrations skill) |249| `find_and_attach_pdfs` | Phase 3/5: Batch OA PDF lookup + auto-attach via Unpaywall |250| `get_item_fulltext` | Phase 3: Read indexed full text + validate post-upload content |251| `add_linked_url_attachment` | Phase 5: Fallback when PDF import fails |252| `inject_citations` / skill | Phase 5: Create live Zotero citations in .docx |253| `add_items_by_doi` | Phase 2: Build bibliography in Zotero |254| `add_items` | Phase 2: Add books, theses, reports, etc. that do not have a DOI |255| None of the above | Ask the user to provide source material directly |256257**Zotero integration**: If the `zotero-mcp-integrations` skill is available and the user258requests PDF upload, follow its Step 2b during Phase 5. For citation injection,259follow its Steps 4-6 during Phase 5.260261---262263## Quality Checklist264265Before delivering the final document, verify:266267- [ ] Every factual claim has a citation268- [ ] Specific numbers (percentages, sample sizes, CIs) are cited — abstracts are a valid source for these269- [ ] Contradictions in the literature are acknowledged270- [ ] Limitations are stated271- [ ] The article answers the question defined in Phase 1272- [ ] Citations are numbered/formatted correctly for the chosen style273- [ ] Abstract-only sources are used appropriately (factual claims only, not interpretive)274- [ ] **Source Transparency disclosure** was presented listing all abstract-only sources275- [ ] If user requested PDF upload: freely available PDFs were verified and uploaded; paywalled sources listed in a table for the user276- [ ] No PDF was uploaded without prior content verification (title, authors, actual content)277278---279280## Common Failure Modes281282| Failure | Cause | Prevention |283|---|---|---|284| Vague claims | Paraphrasing instead of citing specific data | Extract concrete numbers from abstracts or full text |285| Missing context | Skipping Phase 1 scoping | Always define scope before searching |286| Uncritical synthesis | Not considering methods/limitations | When abstract lacks detail, retrieve full text for essential sources |287| Citation errors | Post-hoc citation insertion | Cite while writing, not after |288| Narrow perspective | Too few sources | Search broadly in Phase 2 |289| Hidden abstract-only use | Not disclosing which sources were abstract-only | Always include Source Transparency disclosure |290| Wrong PDF in Zotero | Uploading without content verification | Verify (title+authors+content) → upload → validate |291292---293294## Paywalled Sources295296Many high-impact journals (Lancet, NEJM, Nature, JAMA, Cell, Science) restrict297full-text access to subscribers. This is a normal part of scientific literature,298not a workflow failure.299300**What you CAN do with abstract-only sources (paywalled or otherwise):**301- Cite specific quantitative results reported in the abstract (e.g., "HR 0.67, 95% CI 0.54-0.82")302- Reference the study design and sample size stated in the abstract303- Cite the main conclusion as stated by the authors304305**What you CANNOT do:**306- Describe methodology details not in the abstract307- Discuss subgroup analyses, secondary endpoints, or limitations not mentioned in the abstract308- Use the source as a primary reference for interpretive or mechanistic claims309310**In all cases:** include abstract-only sources in the Source Transparency disclosure.311312---313314## Dependencies315316This skill has no hard dependencies. It works with whatever tools are available.317For optimal results, the following tools enhance the workflow:318319- **PubMed MCP server**: Structured biomedical literature search320- **Zotero MCP server**: Reference management and citation injection321- **web_search / web_fetch**: General source discovery and full-text retrieval322- **docx skill**: Professional document generation323- **pdf skill**: PDF reading and generation324
Full transparency — inspect the skill content before installing.