academic-research-mcp-plugin

# Research Learning Tutor A Claude Code plugin that teaches any subject using only retrieved, cited sources — never parametric knowledge. ## What This Does Instead of teaching from training data (which can hallucinate), this plugin: 1. **Searches** academic papers (Semantic Scholar) and web resources (MIT OCW, 42 School, OSSU, etc.) 2. **Checks** which sources are freely available vs. paywalled 3. **Generates** a citation-grounded learning report 4. **Asks you** when paid resources are needed 5. **Teaches** interactively based only on retrieved evidence ## Persona This is not a cheerful tutor. It is a cold researcher and strict professor: - Facts with citations, no encouragement - Points out errors directly - Verifies your understanding before proceeding - Declares gaps when evidence is insufficient ## Installation ```bash # From GitHub claude --plugin-dir /path/to/research-learning-tutor ``` On first session, dependencies (Python + Node.js) install automatically via SessionStart hook. You will be prompted for: - **Semantic Scholar API key** (free at [semanticscholar.org/product/api](https://www.semanticscholar.org/product/api)) - **Unpaywall email** (free, used for rate limiting) ## Usage ``` /learn transformer attention mechanism ``` The plugin will: 1. Ask what you already know about the topic 2. Search academic databases and web resources 3. Generate a learning report with [FREE]/[PAID] tags 4. Request paid resources if needed 5. Teach interactively with comprehension verification ## Requirements - Claude Code >= 1.0.0 - Python 3.10+ - Node.js 18+ All Python/Node.js packages install automatically. ## Architecture - **research-tutor SKILL** (opus) — core teaching logic, citation enforcement - **researcher agent** (sonnet) — background source discovery, isolated context - **academic-search MCP** (Python FastMCP) — Semantic Scholar + Unpaywall - **web-search MCP** (Python FastMCP) — DuckDuckGo for non-academic sources - **pdf-parser** (Node.js) — PDF to markdown conversion ## Pedagogical Foundation Adapted from: - Google LearnLM: cognitive load management, metacognition - OpenAI Study Mode: prior knowledge assessment, reinforcement - Original: citation enforcement, paid resource flow, no-hallucination guarantee ## License MIT