A research-institute-in-a-server. An MCP server mirroring the skills and workflows of an industry-leading research institute, with local-first server-side retrieval.
Server has 27 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, input schemas are inferred from smoke-test.mjs rather than directly visible in tool registration code. Parameter descriptions exist but are often terse (10-30 chars). Output schemas are not documented. Error handling is minimal, no recovery guidance or actionable error messages visible. Tool composition is strong (each does one thing), but schema completeness and LLM-optimized descriptions are weak. Baseline: 194 chars avg description; this server averages ~80 chars.
Critically appraise a single study and return a structured scorecard. Classifies the design, applies the matching risk-of-bias framework (RoB 2 / ROBINS-I / observational / CASP), assesses the results by magnitude and precision rather than significance alone, rates certainty using GRADE, and judges applicability to a decision context. Feed it an abstract or full text (e.g. retrieved via fetch-url).
Appraise a non-study source (press release, analyst note, legislation, news, blog, official statistic) for credibility and provenance — the companion to appraise-evidence. Places it in the 7-tier source-quality cadence, runs CRAAP and SIFT checks, flags interest/bias, and recommends a weight (High / Moderate / Low / Orientation-only) and a one-line provenance label. Use for grey literature and for novel or fast-moving topics where peer-reviewed evidence is thin.
Assess the ethical dimensions of a study against the Belmont Report principles.
Fetch an author's profile and works by name or ORCID.
Check a study or review against reporting standards (CONSORT, PRISMA, STROBE, etc.).
Tool schemas inferred from smoke-test.mjs, not visible in actual tool registration code (src/tools/*.ts). Cannot verify that registered tools match documented schemas or that input validation is enforced.
Output schemas not documented. Tools return unstructured text or LLM-generated content (e.g., appraise-evidence returns 'structured scorecard', synthesise-evidence returns narrative/tabular/meta-analytic). LLMs cannot plan downstream calls or extract specific fields.
Parameter descriptions are terse (10-30 chars avg). Rubric baseline is 72 chars. Descriptions like 'Search query' or 'Maximum results' lack context for LLM selection. Should explain WHEN to use each parameter and what happens if omitted.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2025-06-18+ | v2 |
Explore the citation graph: works that cite a given work (forward citations) or works it cites (backward citations).
Design a study: choose a design, justify it, and scaffold the protocol.
Develop a measurement instrument (survey scale, interview guide, observation protocol).
Export a bibliography in BibTeX, RIS, or CSL-JSON format.
Extract text and metadata from a PDF URL or file.
Fetch metadata and optionally full text for a scholarly work by DOI, OpenAlex ID, or arXiv ID.
Fetch and parse a URL, returning clean markdown.
Assess the governance and risk profile of a data use case or ML model.
Conduct a structured peer review of a manuscript.
Generate a people-analytics analysis template for a given topic.
Plan the statistical or qualitative analysis for a study.
Scope a research question using the FINER criteria (Feasible, Interesting, Novel, Ethical, Relevant).
Search scholarly literature across OpenAlex, Crossref, arXiv, and OSF preprints.
Synthesise evidence from multiple studies: narrative, tabular, or meta-analytic synthesis.
Translate research findings into actionable recommendations for a specific audience.
Read a note from an Obsidian vault.
Search an Obsidian vault by filename or content.
Write or update a note in an Obsidian vault.
Search the open web for grey literature, press releases, analyst commentary, and regulatory sources.
Generate a workforce survey template for specified constructs.
Save a work to a local Zotero library.
Search a local Zotero library.
No error handling or recovery guidance visible. Tools that call external APIs (fetch-url, web-search, search-literature, fetch-paper) have no documented timeout, rate-limit, or 'not found' behavior. LLMs cannot self-correct on failure.
Vault tools (vault-write-note) accept 'overwrite' boolean with no confirmation step. Agents can silently overwrite notes without user approval. Destructive operations should support dry-run or confirmation.