Synapse Course

SYNAPSE

An accessibility-first, multi-LLM AI tutor for secure software development education.

Learning designed for focused, creative minds.

🔗 synapse-course.com

Live platform · Screencast · Archived release (DOI) · Cite this work · Contributing

---

Evaluating SYNAPSE (for reviewers)

The fastest way to try SYNAPSE is the live instance — no installation required:

• 🔗 Live platform: https://synapse-course.com
• ▶️ Demo video: https://youtu.be/9R17KC47qQI
• 🔑 Reviewer access: registration is free and open

To run it locally instead, see Running locally below. Note that local execution needs your own Anthropic, OpenAI and Google API keys; the live instance above needs none.

Overview

SYNAPSE is a publicly deployed adaptive tutoring platform for Java programming, Python for cybersecurity and forensics, and secure software development. It pairs hands-on secure-coding practice with a neurodivergent-first interface designed for learners with ADHD and executive-function needs, and a multi-LLM orchestration layer that coordinates several language models behind a single pedagogical workflow.

Key features

• Multi-LLM orchestration via MCP — coordinates Claude, GPT-4o, and Gemini through the Model Context Protocol, routing each interaction by pedagogical intent (Socratic guidance, worked examples, conceptual explanation).
• Three-stage Socratic hint policy — withholds direct answers and guides reasoning in progressively more specific stages to reduce over-reliance.
• Neurodivergent-first accessibility layer — always-visible support throughout the platform: dyslexia-friendly fonts, focus mode, idle-aware encouragement, calming audio, and adaptive interface controls.
• ShopSecure — a purpose-built, deliberately vulnerable Flask web application whose exercises map to six OWASP Top 10 (2021) categories, where learners practise the detect–understand–remediate loop on a working system.
• Dual-language code execution — Python runs in a restricted server-side sandbox; Java compiles and runs as an isolated OpenJDK 21 subprocess under resource limits.

Architecture

An asynchronous Python (aiohttp) backend behind an Nginx/Cloudflare reverse proxy, with five interacting subsystems:

• an AI tutor layer coordinated by an MCP server (Claude → GPT-4o → Gemini)
• a course-content engine
• a persistent accessibility layer
• a dual-language code executor
• ShopSecure, running in an isolated Docker container

Persistent storage uses PostgreSQL.

Repository structure

• ai/, ai_clients/ — MCP router and provider clients (Claude, OpenAI, Gemini, DALL·E)
• core/ — tutoring engine, course content, idle/emotion logic
• web/ — HTTP request handlers (auth, course, research, labs, metrics)
• database/ — models and schema
• labs/ — containerised lab environment (Docker session management)
• templates/, static/ — frontend (HTML, JS/CSS, course-content JSON, accessibility assets)
• app.py — application entry point

Running locally

> Requires Python 3.12, PostgreSQL, OpenJDK 21, and Docker (for ShopSecure / labs). You will need your own API keys for Anthropic, OpenAI, and Google.

1. Clone the repository.
2. (Recommended) Create and activate a virtual environment: python -m venv .venv && source .venv/bin/activate — then install dependencies: pip install -r requirements.txt
3. Copy .env.example to .env and fill in your own API keys (Anthropic, OpenAI, Google) and your DATABASE_URL.
4. Create an empty PostgreSQL database (e.g. createdb security_tutor), set DATABASE_URL in .env, then load the schema: python -m database.init_db (applies database/schema.sql, the full table schema)
5. Start the app: python app.py, then open http://localhost:6281 in your browser and register an account.

Optional — only to use the MCP server with Claude Desktop: also copy claude_desktop_config.example.json to claude_desktop_config.json and fill it in. This is not needed to run the web platform.

(OpenJDK 21 and Docker are only needed for the code-execution sandbox and ShopSecure/labs — not to run and explore the tutor.)

Citation

If you use SYNAPSE in your work, please cite the archived software:

> Ferrara, G. (2026). SYNAPSE: A Multi-LLM Orchestrated AI Tutor for Secure Software Development Education. Zenodo. https://doi.org/10.5281/zenodo.20480483

A tool-demonstration paper describing SYNAPSE has been submitted to ICSME 2026.

License

Released under the MIT License — see [LICENSE](LICENSE). © 2026 Giusy Ferrara.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: whitepanther69
• Source: whitepanther69/synapse-course
• License: MIT
• Homepage: https://synapse-course.com

Install and usage instructions live in the source repository linked above.