ed-tech-system-mcp
This server is an MCP (Model Context Protocol) server that provides ed-tech workflows, educational content discovery, AI-assisted authoring, and tutoring capabilities to external clients.
Run liveness checks via
health_check.Search educational YouTube videos with language, result count, and safe-search options.
Retrieve educational documents enriched with complementary videos via
find_documents.Execute a LangGraph document + video discovery workflow with search-term expansion, traceable execution steps, and structured results (
run_workflow).Generate lesson/quiz/project content and research articles through agent workflows.
Author curriculum content end-to-end: search graph nodes, validate lessons/quizzes/projects/test boilerplate, save authored lesson trees to the backend, and scaffold/validate mock tests.
Run a Socratic tutor for interactive teaching turns.
Support project mentor review by collecting review context and producing project reviews.
Integrate with Supabase RPCs, DuckDuckGo/Tavily web search, YouTube Data API v3, LangGraph/LangChain, Redis caching, and backend embedding/RAG services.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ed-tech-system-mcp Find documents about machine learning with related videos"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ed-tech-system-mcp
Domain-Driven MCP (Model Context Protocol) server for ed-tech workflows. The server exposes validated MCP tools backed by LangGraph agents, web search, and YouTube video discovery — all organized with Clean Architecture and DDD. Document embedding and retrieval live in the backend (ed-tech-system-backend embedding service); the MCP server no longer runs RAG.
What this project does
External MCP clients call MCP tools that validate input with Pydantic, delegate to application workflows and LangGraph agents, and reach external systems through domain ports implemented in the infrastructure layer.
MCP tools
Tool | Module | Purpose |
|
| Liveness probe |
|
| Educational YouTube search |
|
| Expand lesson metadata into 4–5 search terms for document/video lookup |
|
| Research article generation workflow |
|
| Lesson/quiz/project content generation |
|
| Graph leaf → generate → validate → save (draft/publish) |
|
| Curriculum graph leaf search |
|
| Content validation |
|
| Persist authored lesson tree via backend RPCs |
|
| Mock assessment scaffolding |
|
| Socratic tutor turn |
|
| Project mentor review |
Handlers use MCP tool caching (when enabled), latency logging, privileged auth where required, and domain error mapping at the protocol boundary.
Integrations
Capability | Integration |
Curriculum graph + authoring RPCs | Supabase RPCs via authoring backend client (anon + manager JWT) |
Web search | Tavily |
Video discovery | YouTube Data API v3 |
Agent orchestration | LangChain / LangGraph |
Caching | Redis when |
Document embedding / RAG | Backend |
Related MCP server: langgraph-clean-arch
Architecture
Clean Architecture under src/mcp_server/. Dependency rule: Domain has no framework I/O; Infrastructure implements Domain ports; Interface and Application depend inward only.
entrypoint → interface → application → domain ← infrastructureLayer | Path | Responsibility |
domain |
| Entities, ports, validators, curriculum enums — no MCP/LangGraph/Supabase |
application |
| LangGraph |
interface |
| MCP tools ( |
infrastructure |
| Supabase, search, YouTube, Groq, Redis, cache adapters |
entrypoint |
| Bootstrap and composition root only |
Changelog folders use the same names plus tests, performance, code-health, refactor — see ARCHITECTURE.md § Changelog layer names.
Read next: ARCHITECTURE.md (layer rules, tree, anti-patterns) · AGENTIC_ARCHITECTURE.md (graphs / tools) · OBSERVABILITY.md (execution traces).
Quick start
Prerequisites
Python 3.12 (see
requires-pythoninpyproject.toml)uv — environment and dependency manager
Doppler CLI (recommended for secrets) or a local gitignored
.env
Install
uv python install 3.12
uv sync --all-groupsConfigure secrets
Secrets never enter git. Use Doppler (team) or a local .env (solo dev).
doppler login
./scripts/doppler/setup-local.sh
./scripts/doppler/bootstrap-from-env-example.sh # first time only — uploads placeholders
# Fill real values in the Doppler dashboard → ed-harness-systemRequired variables: APP_ENV, SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY, YOUTUBE_API_KEY.
Optional: GROQ_API_KEY (only when an LLM path is invoked — lazy-init at first use), TAVILY_API_KEY, LOG_LEVEL (applied at bootstrap via configure_logging()). Staging/production also require CACHE_ENABLED=true and REDIS_URL (local/CI keep the default off).
See ENVIRONMENT_SETUP.md for the full secrets workflow.
Run the MCP server
# With Doppler
doppler run -- uv run mcp-server
# With local .env (APP_ENV=development)
uv run mcp-serverInspect traces in tests
Traces are captured programmatically by workflow_trace.py and workflow_llm_trace.py and asserted in pytest. See OBSERVABILITY.md for trace field details and debugging patterns.
Development
Day-to-day commands
uv sync --frozen # after pulling lockfile changes
uv run mcp-server # start server
uv run ruff check src/ # lint
uv run ruff format --check src/
uv run mypy src/ # type check
uv run pytest # tests (143 cases as of 2026-07-21)Engineering backlog
Audit findings are triaged into backlog/BACKLOG.md (RICE-ranked, traceable to changelog audits). As of 2026-07-21: 23 done, 6 deferred (adapter HTTP implementation, profiling, trace IDs).
Add dependencies
uv add some-package # runtime
uv add --group dev some-tool # dev onlyDo not use pip install in this repo — it bypasses the lockfile.
Quality gates (CI parity)
uv sync --frozen --all-groups
uv run ruff check src/
uv run mypy src/
npm run lint:architecture # layer imports + boundary patterns (also runs on git push)
uv run pytestGit hooks: Husky pre-commit runs public-repo safety checks (sensitive files, tracked leaks, secret scanners); pre-push re-checks tracked safety, scans pushed commits for secret content, then runs architecture lint — neither blocks the other tier.
Run quality-gate commands from the repository root (ed-tech-system-mcp/), not ui/. The same scripts are also available inside ui/ via npm run hooks:test and npm run lint:architecture.
Project layout
.
├── src/mcp_server/
│ ├── domain/ # Entities, ports, validators, enums
│ ├── application/ # Agents, runners, LLM routing, authoring services
│ │ └── agents/ # LangGraph packages (content_generation, socratic, …)
│ ├── interface/ # MCP tools + validation
│ ├── infrastructure/ # Adapters (search, video, LLM, cache, clients)
│ ├── wiring.py # Composition root
│ ├── settings.py
│ └── main.py # mcp-server entry
├── tests/ # pytest + architecture lint
├── changelog/ # Agent memory: {DATE}/{LAYER}/ (local)
├── scripts/ # Doppler, hooks, Render, dev helpers
├── docs/assets/ # README screenshots
├── ARCHITECTURE.md # Layer boundaries (canonical)
├── AGENTIC_ARCHITECTURE.md # Agent graphs and tool orchestration
├── OBSERVABILITY.md # Workflow UI, trace replay
└── ENVIRONMENT_SETUP.md # uv, secrets, CI, MCP client configDocumentation index
See the documentation matrix at the end of this file for canonical docs and changelog artifacts.
MCP client integration
Register the server in your MCP host using the project interpreter:
{
"mcpServers": {
"ed-tech-system": {
"command": "doppler",
"args": ["run", "--", "uv", "--directory", "/absolute/path/to/ed-tech-system-mcp", "run", "mcp-server"]
}
}
}Alternative patterns (local .env, uv launcher) are in ENVIRONMENT_SETUP.md § MCP client integration.
Documentation matrix
Read the minimum doc set for your task. Do not load everything.
Canonical docs (repo root)
Document | Read when |
First visit — overview, quick start, MCP tools | |
Any code change — layers, ports/adapters, deps per layer, file layout, anti-patterns | |
LangGraph/LangChain agents, LLM wiring, tool taxonomy, DB/web/video flows | |
Execution traces and debugging | |
|
Conflict resolution: ARCHITECTURE.md wins on layer boundaries; AGENTIC_ARCHITECTURE.md wins on orchestration semantics.
Engineering backlog
Document | Read when |
RICE-ranked tasks from audits; status tracking | |
Scoring rubric and priority formula for backlog items |
Changelog memory (changelog/{DATE}/{LAYER}/)
Local engineering memory (often gitignored). {LAYER} must match an architecture layer or audit folder:
domain · application · interface · infrastructure · entrypoint · tests · performance · code-health · refactor
File pattern | Purpose |
| Scope and gaps before coding |
| Execution checklist and status |
| Pre-merge review findings |
| Behavior catalog before writing tests |
| Coverage verdict after tests pass |
| Performance bottleneck findings |
| Maintainability / dead-code findings |
| Merged refactor actions from audits |
| Recursive-loop parameters and iteration log |
Pairing: IMPLEMENTATION{N} ↔ INVESTIGATION{N}; CODE_REVIEW{N} ↔ same {N}. Full protocol: .cursor/rules/changelog-agent-memory.mdc.
Quick routing
Code in a layer? → ARCHITECTURE.md (+ AGENTIC_ARCHITECTURE.md if agents/tools/LLM)
Traces / debugging? → OBSERVABILITY.md
Environment / CI? → ENVIRONMENT_SETUP.md
Secrets / Doppler? → ENVIRONMENT_SETUP.md § Secrets & safety
Tests / merge gate? → pytest + quality gates in ENVIRONMENT_SETUP.md
Audits / cleanup? → backlog/BACKLOG.mdLicense
Available Tools
4 toolsfind_documentsB
Retrieve educational documents enriched with complementary videos.
Inputs: query (required), video_limit, document_limit. Returns: Validated output for document + video discovery tool calls.
health_checkA
Verify the MCP server is running.
Inputs: no named parameters. Returns: result
run_workflowC
Execute the document + video discovery LangGraph workflow.
Inputs: query (required), video_limit, document_limit. Returns: Validated output for LangGraph workflow execution.
search_youtubeC
Search for educational YouTube videos matching a query.
Inputs: query (required), language, max_results, safe_search. Returns: Validated output for video search tool calls.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 4 tools
Tools are distinct: health_check is server status, search_youtube targets only videos, find_documents focuses on documents with videos, and run_workflow executes a combined workflow. Minimal overlap.
All tools use snake_case with a verb_noun pattern, though health_check is noun_verb. The pattern is consistent and readable.
With 4 tools, the server feels too sparse for an 'ed-tech system.' The core functionality (search, documents, workflow) is minimal, leaving many expected operations uncovered.
The server lacks CRUD operations for resources, user management, and other typical ed-tech features. The tool set is focused on discovery only, which is incomplete for the broad server name.
Related MCP Connectors
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Unified 15-tool SEO, SERP, clean Markdown & GEO (LLM citation Share-of-Voice) MCP server.
MCP server for Clipkit — gives AI agents a video toolbox via the Clipkit schema.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA production-grade MCP server designed for multi-tenant, authenticated, and observable AI agent systems, enabling secure tool execution across heterogeneous data sources.68MIT
- FlicenseNot gradedqualityDmaintenanceMCP server implementing clean architecture with LangGraph for building and managing agent workflows.-
- FlicenseNot gradedqualityDmaintenanceMCP server for The Culture, a fashion-focused social media app, exposing Supabase database operations and Hugging Face model inference as tools for use in Claude Code, n8n agents, and the Larry orchestrator.-
- FlicenseAqualityDmaintenanceReal TypeScript MCP server for connecting Zynx agents to Supabase through safe, allowlisted tools.81-