# Lit-Review-Council MCP Server > Multi-Agent Literature Review Council MCP Server **Status:** Measured to an artifact. **Role:** Author / Google × Kaggle AI Agents Intensive. **Timeline:** Jun 2026. **Category:** Agentic AI & MCP. Multi-agent research synthesis engine and Model Context Protocol (MCP) server that orchestrates parallel academic and practitioner research with Borda-count consensus. ## Measurements | Value | What it measures | | --- | --- | | PyPI | published, install via uvx lit-review-council; listed on the MCP Registry | | 3 | independent reviewer agents ranked by Borda count | | 2 | research tracks per topic: academic (ArXiv/OpenAlex) and practitioner (GitHub) | ## The artifact that settles this Consensus is a counted Borda tally over named reviewers, not a model asked to agree with itself. Quoted from `Ravicha2/lit-review-council` at `scoring.py#L12-L35` (40c4c19), `src/scoring.py#L12-L35`, verbatim: > def tally_ensemble_rankings(rankings: List[Optional[Any]]) -> Tuple[str, List[str]]: > """ > Applies Borda count (2 pts for 1st, 1 pt for 2nd) and returns the winning label > and the list of rationale strings. > """ > scores = {"A": 0, "B": 0} > reasons = [] > > reviewers = ["Researcher", "Engineer", "Architect"] > > for i, r_obj in enumerate(rankings): > reviewer_name = reviewers[i] if i < len(reviewers) else f"Reviewer_{i}" > > if r_obj: > lst = r_obj.ranking if hasattr(r_obj, "ranking") else r_obj.get("ranking", []) > rat = r_obj.rationale if hasattr(r_obj, "rationale") else r_obj.get("rationale", "") > if lst and len(lst) >= 1: > # 1st place gets 2 points, 2nd gets 1 point > if lst[0] in scores: scores[lst[0]] += 2 > if len(lst) > 1 and lst[1] in scores: scores[lst[1]] += 1 > reasons.append(f"{reviewer_name} rationale: {rat}") > > top_label = "A" if scores["A"] >= scores["B"] else "B" > return top_label, reasons Permalink: ## The intuition Deliver rigorous, multi-perspective literature synthesis with verifiable academic and open-source citations directly in developer IDEs. - The spark: Single-prompt LLM research summaries suffer from confirmation bias, fabricated citations, and an inability to balance theoretical academic papers with practical open-source production implementations. - The naive failure mode: Single-agent LLMs hallucinate non-existent academic references, accept unvalidated preprint claims, and lack diverse perspectives when assessing the feasibility of architectural patterns. ## The problem Citation hallucination, single-prompt bias, and IDE context limits in automated research synthesis. **Constraints** - Strict Model Context Protocol (MCP) compliance for universal integration into Claude Code, VS Code, and Cursor - Zero hallucinated or unreachable citations permitted in final synthesis output **Edge cases handled** - LLMs generating plausible-sounding but fictitious DOI and ArXiv identifiers - Preprint source bias skewing recommendations toward unproven theoretical approaches - IDE context window overflow when ingesting large volumes of paper abstracts ## Why it was built this way Dual-track research agents, 3-reviewer Borda ranking, programmatic citation validation, and MCP protocol distribution. Dual-track parallel research waves paired with an independent 3-reviewer Borda-count scoring council and rigorous programmatic citation verification. **Trade-offs** - Dual-Track Research (Academic + Practitioner) — Dispatches specialized agents to query ArXiv/OpenAlex for theoretical rigor while simultaneously querying GitHub/PyPI for production adoption and practical ergonomics. (against: Single source search) - Borda-Count Consensus Ensemble — Three independent reviewers (Researcher, Engineer, Neutral) evaluate anonymized reports on novelty, methodology, and practicality before aggregating ranks via Borda voting. (against: Single reviewer scoring) - MCP Protocol Server Distribution — Packaging as an MCP server installable via uvx enables developers and AI coding agents to trigger comprehensive literature reviews from their command-line and IDEs. (against: Web dashboard only) **Guardrails** - Programmatic API verification verifying that all cited DOIs and ArXiv IDs exist and match paper titles - Dangling reference rejection filter pruning unvalidated citations from final reports - Source-tier classification giving higher confidence weight to peer-reviewed publications ## Outcomes Distributed via PyPI and MCP, providing verified multi-agent literature syntheses directly in developer IDEs. **How it was verified** - Published to PyPI (installable via uvx lit-review-council) and listed on the MCP Registry, with documented install paths for Claude Code and VS Code. - Every URL in a finished report must appear in the source references or the synthesis step retries, so dangling citations are rejected before output rather than reviewed after it. **Impact** - Lets researchers and engineers run a multi-perspective literature review from their IDE. - Demonstrated effective multi-agent consensus mechanisms for reducing LLM cognitive bias. **Takeaway:** Multi-agent consensus coupled with programmatic deterministic validation solves the hallucination problem in complex research synthesis. ## Stack Google ADK, Python, MCP, OpenAlex API, ArXiv API, uv, Claude Code ## References - github: - pypi: --- ## Fetching the rest of this site - This page as HTML: - Concise index: - Full dossier: - When to use this material: - API description (OpenAPI 3.1): - API reference: - Sitemap: