Story Intelligence

The engine, headless, in your stack. Your users, your brand, our narrative and safety machinery underneath. This catalog is deliberately capability-shaped, not a subset of Storytailor’s account CRUD.

Endpoint & auth

POST https://mcp.storyintelligence.dev/mcp
Authorization: Bearer si_live_<keyId>_<secret>

Organisation-scoped keys, stored hashed, compared in constant time. Each tenant executes against an isolated service principal, so one partner’s key can never reach another’s data. initialize and ping work unauthenticated so a client can complete a handshake and read a proper auth error.

Capabilities

Area	Tools
Safety	`safety.classify`
Narrative	`narrative.calibrate`, `story.plan`, `story.compose`, `story.status`, `story.continue`
Character	`character.define`, `character.render`, `character.status`, `character.continuity`, `character.adapt`
Delivery	`narration.synthesize`, `narration.timing`, `content.localize`, `kid.interpret`
Vetted tier	`therapeutic.support`, `emotion.record`, `emotion.patterns`, `insights.findings`

Vetted-tier tools do not appear in tools/list unless your organisation has been vetted for them. Tiering is by limits, never capability. Every tier gets the full engine and the full safety layer.

A first generation

curl -sX POST https://mcp.storyintelligence.dev/mcp \
  -H 'content-type: application/json' \
  -H "authorization: Bearer $SI_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
        "name":"story.compose",
        "arguments":{"storySeed":"A curious fox follows fireflies to a meadow concert",
                     "readingAge":"6-8","language":"en"}}}'

Returns a story id immediately; poll story.status until prose, cover, scene art and narration report ready. Text lands in about 30 seconds, a complete story in roughly three minutes.

Safety is not optional

Verdicts are computed server-side and cannot be disabled by any flag, tier, or key. They fail closed: an internal error returns block, never unchecked content. Every verdict ships the evidence behind it, never a bare score. safety.classify is effectively free within generous caps, on purpose: we will not create an incentive to skip a safety check.

The diagnostic lock

Story Intelligence never diagnoses. Emotion and insight tools surface what a child actually expressed, with the material it came from, so a qualified human can interpret it, using observed vocabulary only, never a score, flag, or risk rating. The human diagnoses; the product observes. This is binding on every tier and every future tool.

Limits & spend

Every organisation has a per-minute rate limit and daily call and generation ceilings. Exceeding one returns a clear error with a retry hint, never a silent truncation. A “generation” is one finished story or character; planning, status polls, calibration and safety checks are plain calls.