What you can build on

Kontaim exposes its activity engine through a Model Context Protocol (MCP) server. Any MCP client — Claude, ChatGPT, or your own agent stack — can search the public Templates, generate a brand-new interactive experience from a plain-text description, and hand back a playable link. Activities auto-grade themselves, run on any browser (Chromebooks included), and require no student accounts.

How an integration works

Create — call create_activity with a teacher-style description. You get an activity token back immediately; the AI builds in the background (typically 1–3 minutes).
Poll — call check_activity with the token until the build is ready.
Play — the result is a hosted play link (https://kontaim.com/made/<token>). Anyone with the link can play instantly — no login.
Claim — from the play page, the creator can save the experience into a free Kontaim account to edit it, assign it to a group, and track results.

The token is the capability: whoever holds it can view and claim that experience. Kontaim stores no personal information about the person who requested the build.

Connect your client

The MCP endpoint is:

https://kontaim.com/api/mcp

Claude — Settings → Connectors → Add custom connector, paste the endpoint URL.
ChatGPT — enable Developer mode in settings, then add the endpoint as a connector. (Read-only tools work on all plans; creating activities from a custom connector requires a Business/Edu/Enterprise workspace until the app is listed in the directory.)
Your own agent — every major framework speaks MCP. With the Vercel AI SDK, point an MCP client at the endpoint and the three tools below become callable tools in your loop.

Or speak JSON-RPC directly:

curl -X POST https://kontaim.com/api/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "create_activity",
      "arguments": {
        "description": "Grade 6 Science: Ecosystems. Build and maintain a balanced ecosystem by introducing species across 3 biome levels.",
        "grade": "6",
        "subject": "science"
      }
    }
  }'

Authentication

The endpoint works without authentication for evaluation, with the preview limits below. Partners get an API key that lifts those limits to a dedicated quota, attributes every build for usage reporting, and unlocks partner features like QA-reviewed builds. Send it as a bearer token:

Authorization: Bearer ak_live_...

Keys are issued during partner onboarding (contact us), shown once at creation, and revocable at any time. Treat them like passwords — server-side only, never in client-side code.

Tools

Tool	Arguments	Returns
`search_library`	`query?`, `subject?`, `grade?`	Up to 8 published experiences with view-and-play links. Free, unlimited.
`create_activity`	`description` (required), `grade?`, `subject?`, `title?`, `quality?`	An activity token, immediately. The build runs in the background for 1–3 minutes.
`check_activity`	`activity_token`	Build status; when ready, the hosted play link.

Limits (preview): 3 builds per day per user, with a short cooldown between builds and a platform-wide daily ceiling. Partner API keys replace all of these with the partner's own quota.

QA-reviewed builds (partners)

With an API key, quality: "reviewed" runs an expert review after the build — the same classroom-ready rubric used by our internal content pipeline (works end-to-end, genuinely interactive, accessible, age-appropriate, meaningful grading) — and applies one targeted fix pass when the review finds blocking issues. It adds roughly 1–2 minutes and is billed separately from the build.check_activity reports the QA outcome alongside the play link.

Prompts that work well

“Make an interactive fractions game for my 3rd graders that I can assign tomorrow — no timers, mastery over speed.”
“Find me a ready-to-play ecosystems activity for grade 6 science.”
“Build a vocabulary sorting activity for my ESL students on weather words, grade 4.”

Data & ownership

Zero PII by design. Builds are keyed to anonymous tokens. When ChatGPT calls the server it sends an anonymized per-user id used only for rate limiting; Kontaim never receives names or emails through this surface.
Your users own what they make. An unclaimed experience belongs to nobody; claiming it moves it into the creator's own Kontaim account.
Privacy posture is documented in the Trust Center, Privacy Policy, and compliance matrix.

On the roadmap for partners

For platforms that want deeper integration than an assistant connector, the partner program adds (in active development — reach out to join the design phase):

REST API (/api/v1) as a conventional alternative to MCP
Embeddable runner — render experiences inside your product via iframe, with real-time completion events posted to your page
Grade delivery — signed webhooks on completion, xAPI statements, and direct LRS forwarding
Prompt flavor — a partner-level style preamble so generated experiences match your voice
SCORM export for legacy LMS environments

Building a learning platform, an L&D product, or curriculum tooling and want experiences as infrastructure? Get in touch — we're onboarding a small number of design partners.

Build with Kontaim