mcp-ai-tutor

An MCP server that exposes AI Tutor Card declarations as tools. Drop into Claude Desktop, Cursor, or any MCP-compatible client. The agent — or the procurement reviewer driving it — gains six tools for inspecting any AI tutor's audience, pedagogy, safety posture, and FERPA / COPPA / GDPR compliance.

This is the EdTech-flavored sibling of mcp-aeo-server. Same Tool Card discipline, different spec.

Tools

Tool	What it does
tutor_card_well_known_url	Compute /.well-known/tutors/<tutor_id>.json for a vendor origin
tutor_card_fetch	Fetch a published Tutor Card from any URL
tutor_card_validate	Schema-validate an inline document; also runs the COPPA conditional check
tutor_card_inspect	Procurement-grade summary: audience, pedagogy, safety, privacy posture, eval count
tutor_card_subject_check	Given a topic query, classify it as primary, included, excluded, or unknown
tutor_card_coppa_check	Verify the spec's conditional COPPA rule (age_range_min < 13 ⇒ coppa_compliant)

Six total. Tools that act on a document accept either url (the server fetches it) or document_json (an inline JSON string).

Install

npm install -g @mizcausevic-dev/mcp-ai-tutor

Or run without installing via npx:

npx @mizcausevic-dev/mcp-ai-tutor

Claude Desktop config

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "ai-tutor": {
      "command": "npx",
      "args": ["-y", "@mizcausevic-dev/mcp-ai-tutor"]
    }
  }
}

Restart Claude. All six tutor_card_* tools appear.

Workflows this unlocks

District procurement review. Pull two tutors' cards in one conversation and let Claude compare them:

"Fetch both https://vendor-a.example/.well-known/tutors/math-tutor.json and https://vendor-b.example/.well-known/tutors/math-tutor.json, run tutor_card_inspect on each, and tell me which has stricter content filters and which keeps less student data."

Procurement red-team. A board reviewer can ask Claude to surface deal-breakers:

"For tutor.example.com/k12-math, run tutor_card_coppa_check. If it fails, that's a blocker for our K-5 deployment."

Curriculum-to-tutor matching. A curriculum director can map lessons to candidate tutors:

"My middle-school unit covers ratios, proportions, and pre-algebra. For each candidate tutor card I'll pass you, run tutor_card_subject_check on each of those three topics and tell me which tutor covers all three as primary or included."

Sample output: tutor_card_inspect

{
  "tutor_card_version": "0.1",
  "tutor": {
    "id": "kineticgain-k12-math-tutor",
    "name": "Kinetic Gain K-12 Math Tutor",
    "version": "1.4.0",
    "provider": "Kinetic Gain Edu"
  },
  "audience": {
    "ages": "5-18",
    "grades": "K-12",
    "languages": ["en", "es"]
  },
  "pedagogy": {
    "approach": "socratic",
    "homework_policy": "guide_only",
    "assessment_policy": "refuse"
  },
  "subject_scope": { "primary": ["Math"], "includes": 11, "excludes": 4 },
  "safety": {
    "content_filter_strength": "strict",
    "mandated_reporter_protocol": true,
    "human_in_loop_categories": ["mental_health_disclosure", "abuse_disclosure", "self_harm_disclosure", "bullying_disclosure"]
  },
  "privacy": {
    "ferpa_compliant": true,
    "coppa_compliant": true,
    "gdpr_compliant": true,
    "retention_days": 90,
    "parents_see": "summaries_only",
    "school_sees": "summaries_only",
    "shares_with_third_parties": false
  },
  "agent_card_uri": "https://edu.kineticgain.com/.well-known/agents/k12-math-tutor.json",
  "evaluation_count": 3,
  "coppa_check": { "ok": true, "reason": "Audience min age is 5 (<13) and coppa_compliant=true, as required by the spec.", "age_range_min": 5, "coppa_compliant": true }
}

Tests

17 unit tests covering every tool, every classification path of tutor_card_subject_check, every branch of tutor_card_coppa_check, and a published-card fetch via an in-process Node HTTP server (no external network).

npm install
npm run typecheck
npm test
npm run build

Why this beats a vendor brochure

A PDF whitepaper says "we are FERPA-compliant." A signed Tutor Card the buyer's MCP-equipped review agent can read says exactly which fields, in a format two competing tutors can be diffed against. The procurement workflow gets a CI gate, not a sales call.

Compatibility

• Node 18+
• @modelcontextprotocol/sdk ^1.0
• Tested with Claude Desktop, Cursor, and any client speaking MCP over stdio

License

AGPL-3.0.

Kinetic Gain Protocol Suite

Spec	This server's role
AEO Protocol	covered by mcp-aeo-server and mcp-kinetic-gain
Prompt Provenance	covered by mcp-kinetic-gain
Agent Cards	covered by mcp-kinetic-gain; Tutor Cards specialize this spec for education
AI Evidence Format	covered by mcp-kinetic-gain
MCP Tool Cards	covered by mcp-kinetic-gain
AI Tutor Cards	mcp-ai-tutor (this)

---

Connect: LinkedIn · Kinetic Gain · Medium · Skills