Generate clinical protocols for medical devices or drugs with Drug (TPMW001/ICH-GCP) and Medical Device (TPMW045/ISO 14155 + MDR Annex XV) pathways. v1.4.1 elevates three first-class section properties: (1) per-section authority + whole-corpus information gathering (primary source wins on conflict; agent freely reads all corpus documents for supplementary context); (2) verbatim copy-paste mapped dynamically from synopsis-block keywords (§4/§4.1, §5.1/§5.1.1/§5.1.2, §6.1/§6.2/§6.3, §5.5/§7.4, §8.1.1/§8.1.2 byte-for-byte from synopsis when matching block found); (3) figure placeholders ([FIGURE: ...] or [MISSING FIGURE: ...]) for §3.1/§3.2 drug + §2 device. Design-feature existence (SRC/DSMB/iDMC/interim_analysis/sentinel_dosing/adaptive_design/cohort_progression_gate/dose_escalation) is now a first-class lock dimension determined by agent reasoning (study characteristics + corpus mentions + confidence level), not regex grep. Strict identifier taxonomy: sponsor_study_number (canonical), cro_internal_code (NEVER in prose), regulatory_ids (pending|assigned status). Sponsor/CRO entity names always from lock, never skill default; multi-jurisdiction studies flag for sponsor confirmation. ICH E6 bumped to R3 via central regulatory-standards-registry.json (one place to update all citations). Rule-based fact-checker framework (scripts/checkers/*.py): outdated_citations, design_features, cro_internal_code_leakage, software_versions, reference_dedup, broken_cross_refs, hardcoded_org_names. §17 = REFERENCES, §18 = APPENDICES (no MDCE-specific administrative section). export_docx.py: fenced code blocks as monospace, all [MISSING|FIGURE|NOTE|DEVIATION:] render red bold, body-start detector accepts h1-h3 numbered headings, Word heading-level mapping shifted -1 so top sections render as 'X.' not '1.X', scope-label headings skipped. Phase 0 intake auto-detects drug/device, silent defaults for language/terminology.
Convert a single document (PDF / DOCX / PPTX / XLSX / HTML / etc.) into a cleaned markdown bundle. Native text and structure are extracted with docling locally; scanned PDF pages and embedded images are processed by qwen3.6-flash through the Euris model gateway. Output is a folder (or zip) with output.md plus an attachments/ subfolder.
Requires EURIS_API_KEY or ~/.euris/config.toml. The desktop app writes both automatically after login.
Ask life-sciences research questions answered by a backend agent that calls 16 authoritative data sources (PubMed, Europe PMC, ClinicalTrials.gov, openFDA, RxNorm, NCBI Gene/SNP/Protein, AccessGUDID, EUDAMED, NIH RePORTER, CMS Coverage, J-STAGE, KoreaMed, Tavily Web, Financial Modeling Prep). Every answer is written to a .md file with inline [N] citations and structured references — the host agent reports the file path, never echoes the answer. Single ask is a 1-item batch, no conversation cache.
Requires EURIS_API_KEY or ~/.euris/config.toml. The desktop app writes both automatically after login.
Connect to Scite.ai for citation context analysis, evidence-backed literature search, and Smart Citations across 1B+ scientific references. Authenticates via OAuth on first use.
Read and write the user's Microsoft 365 data through Microsoft Graph: Outlook mail, calendar (incl. shared + scheduling assistant), SharePoint sites + lists, OneDrive files (read + write + binary download), Teams chats / channels / messages, online meetings + recordings + transcripts, OneNote, To Do + Planner, contacts, presence, directory lookup. Acquires its own Microsoft Graph token via incremental consent at install time (no overlap with the user's base Euris sign-in). Schema is split into per-module markdown docs that the LLM reads on demand, keeping context usage low while supporting the full Graph API surface.
Requires a real Microsoft Graph token from the Euris desktop app. Do not use a Euris key for Graph.
Read Odoo sales quotations and update unit quantity, hours, and quantity fields on quote lines. Access is scoped to the mcp_odoo_quotation Odoo service account with model/field-level filtering and value validation enforced at both the MCP layer and inside Odoo. Dev environment only (dev.mdceai.com, database mdceai_04_22). Uses MCP streamable-http transport.
Self-contained, agent-driven clinical study protocol generator (drug / ICH-GCP / TPMW001). Materials-first: reads and saves the sponsor materials already in the workspace, builds a frozen design lock by extracting every field the materials answer (verbatim for sponsor-required facts) and asking ONLY for what is genuinely missing, then generates the protocol SECTION BY SECTION into a resumable run directory — each segment its own file, so it can be reviewed, revised and continued. First-class source-priority (per-section primary/secondary, primary wins, verbatim where required). No scripts, no external engine — the agent does everything from the skill prompt. Generation is constrained by anti-hallucination / output-discipline / template-fidelity rules, but the finished protocol is NOT self-reviewed — auditing the output is a separate review skill's job. v0.1.0 covers the drug pathway only; review/QC and the device pathway are out of scope here.