Loading ToolBench
MCP (Model Context Protocol) server for Science AI Journal tools — Pre-Check, AI Review, Research Gaps, Journal Recommender, Duplicate Publication Checker, Article Writer. Use them inside Claude Desktop, Cursor, Cline, or any MCP-enabled client.
Seven tools with complete input schemas and descriptions. Naming follows verb_noun convention (ai_review_paper, start_writer_pipeline, get_writer_pipeline_status, check_duplicate_publication, recommend_journals, pre_check_paper, find_research_gaps). Descriptions are detailed (150 - 400 chars) and explain WHEN to use each tool. Parameters are typed with enums where appropriate (agentType, section, sort, filters). However, output schemas are not documented in the source, LLMs cannot see what fields to expect from responses. Error handling is minimal; no recovery guidance or actionable error messages visible. Tool composition is sound (each does one thing), but some tools lack parameter descriptions for optional fields (e.g., 'language' in start_writer_pipeline has no description of what values are valid beyond 'python | r | matlab | ...').
Run one of the specialist reviewer agents (methodology, language, figures, plagiarism, or literature) on a prepared agent prompt and return the structured editorial decision: score (1-10), verdict (Accept / Minor Revision / Major Revision / Reject), summary, strengths, concerns, detailed multi-section review, confidence band, and questions for the authors. Uses Science AI Journal credits. For the full 5-agent + synthesis flow, prefer the web UI at scienceaijournal.com/ai-review.
Pre-submission duplicate-publication and salami-slicing check. Cross-references the title + abstract against CrossRef, arXiv, medRxiv, bioRxiv, Unpaywall, and a 900k-paper institutional library in ~30 seconds. Returns a status (likely_published, uncertain, or not_found) with a confidence score and message. Free. Use before submission to catch accidental duplicates or to confirm a preprint hasn't been formally published yet.
Surface research gaps + the most-cited and most-recent papers around a query. Returns cross-paper synthesis gaps first (LLM-derived, grounded in ≥ 2 papers), then catalogue-level single-paper gaps. Plus a field overview and ≤ 50 top-cited and ≤ 50 most-recent papers. Uses Science AI Journal credits; rate-limited to 10 requests/hour per IP. Use for early-stage research discovery, literature gap identification, and proposal scoping.
Return the latest writer-pipeline job for a (sessionId, section). Use after `start_writer_pipeline` and poll every 5-10 seconds. Terminal statuses: 'done', 'failed', 'aborted'.
Output schemas not documented. LLMs cannot see what fields responses contain (e.g., does ai_review_paper return 'verdict' or 'decision'? Is it a string or enum?). This forces agents to guess field names and wastes tokens on parsing unstructured output.
Error handling lacks recovery guidance. No visible error messages that tell LLMs what to do next (e.g., 'API rate limit exceeded. Retry in 60 seconds.' or 'Session not found. Create one via the web UI first.'). Agents cannot self-correct.
Parameter 'language' in start_writer_pipeline lacks a description. The inline comment 'python | r | matlab | ...' is not a formal parameter description. LLMs cannot parse inline comments in schemas.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Run a free, zero-LLM-cost pre-submission scoring of an academic paper. Returns predicted publication tier (Tier 1, 2, or 3 probability), the detected research field, and a confidence band. Backed by a local 33,000-paper library via FTS5 BM25. Use this when the user wants a fast sanity-check on whether a paper is ready to submit, or which tier of journal to target. Sub-second response. Free.
Recommend ranked target journals for a paper from a ~1,200-venue index. Each result includes a letter grade (A-F), a match percentage, tier (1-3), publisher, open-access status, 2-year mean citedness, predatory-journal flag, and 2-3 example similar papers that landed at that venue. Free — runs locally via FTS5 + topic-RAG, no LLM call. Use when the user asks 'where should I submit this paper' or wants to compare target venues before deciding.
Enqueue a writer-pipeline job for an existing WriterSession. Returns immediately with a jobId; the orchestrator-worker claims it within ~5 seconds. Default section is step_7_5 — the auto-chain that replaces the three buttons in Step 7.5. Use `get_writer_pipeline_status` to poll until status is 'done', 'failed', or 'aborted'. Uses Science AI Journal credits.
recommend_journals 'filters' object has nested properties but no description of what each filter does (e.g., what does 'minTier' actually filter?). Nested objects without field-level docs are error-prone.
No pagination or result-limiting guidance visible in tool descriptions. recommend_journals caps at 25 results, but users don't know this limit exists. Large result sets risk context window exhaustion.