Vahtian MCP Server

vahtian

Company site for Vahtian — research citation integrity.

Vahtian checks whether each manuscript claim is supported by the evidence cited for it — through a blinded human → AI → adjudication workflow with a decision-gated, undoable write and a hash-chained audit trail.

Vahtian does not decide scientific truth. It records whether a cited source supports a specific claim: who rated it, what the AI advised, how disagreements were adjudicated, and what was finally written.

Human-first. AI-second. Auditable. Free local tools for researchers; infrastructure for journals, guideline groups, and research organizations.

The product ladder

Start — check the claims in your own manuscript

• CiteVahti — check that each claim is supported by the source cited for it (free beta): https://github.com/heidihelena/citevahti

Prepare — get the study and its evidence right before analysis

• StudyVahti — plan a defensible study before data collection (free · live): /studyvahti
• MethodVahti — justify your sample size → COREQ/SRQR methods PDF (free beta for individuals, project licence for institutions): /methodvahti
• DictVahti — clean the REDCap data dictionary before analysis (free · live): /dictvahti
• MatchVahti-Lite — capture citation-worthy evidence from abstracts (free · live): /matchvahti-lite

Scale — systematic-review infrastructure

• ReviewVahti — inter-rater reliability for screening (Cohen's κ / Krippendorff's α) (live): /reviewvahti
• ExtractVahti — DTA data extraction + risk-of-bias (QUADAS-2/-C, PROBAST-AI) → tidy CSV + RoB traffic-light SVG for R; multi-extractor reconcile (κ / α); sealed agent aid (early access): /extractvahti
• GuidelineVahti — panel consensus, RAND/UCLA + GRADE (early access): /guidelinevahti
• AtlasVahti — de-identified validation-pattern analytics (roadmap)

Static pages (one index.html per product, plus 404.html and sitemap.xml), deployed via Cloudflare to vahtian.com. No trackers, no external dependencies.

vahti (Finnish) = sentinel / guard — the watch over what gets cited.

Brand & social tooling

• brand/STYLE.md — the canonical brand spec; brand/PHILOSOPHY.md — the mark system.
• brand/cards/ — 1200×630 Open Graph cards, one per page. Regenerate with npm i @resvg/resvg-js && node brand/cards/generate.mjs.
• Screenshot-led product covers are generated from their real interface previews with node brand/cards/generate-product-covers.mjs (uses the bundled preview browser).
• brand/carousel/ — self-contained LinkedIn carousel builder (noindex, zero external requests). Preset decks per product, theme & format switching, post-caption drafts, deck save/load as JSON, export as PDF (LinkedIn document post) or per-slide PNGs.

MethodVahti package

methodvahti/ is a self-contained Python package (Apache-2.0) for qualitative heterogeneity scoring + a COREQ/SRQR methods-report PDF. See methodvahti/README.md. Tests: cd methodvahti && python -m pytest tests/ -v.