ScryMaster Docs

Build a story with your own AI

You don’t have to click through every scene by hand. Bring an AI you already use — Claude, Gemini, or ChatGPT — describe the adventure you want, and let it draft the whole branching story for you. Then you read it over, tweak the bits you care about, and publish. Free, for everyone, forever.

What’s an “MCP,” in plain English? MCP (Model Context Protocol) is just a standard way to hand your AI a small set of safe buttons for a specific app. Instead of you copy-pasting back and forth, your AI talks to ScryMaster directly: “show me what I’m allowed to do,” “add this scene,” “is this story complete?” Think of it as giving your assistant the keys to the Studio — but a keyring that only opens the doors we choose.

Why this is faster (and safer) than a chatbot DM

  • Describe, don’t click. “A three-scene haunted-chapel heist for level 3, with a trap, a bribe option, and two endings” becomes a real, editable adventure — not a wall of text.
  • It can’t make things up. Your AI is handed the exact list of things the engine can do (the closed action vocabulary). If it invents a command that isn’t real, ScryMaster simply rejects it — so AI-built stories stay valid instead of quietly breaking.
  • The dice stay real. The AI builds the story once; the game engine rolls the dice at play time. That’s why your published adventure runs free, offline, and the same way every time — no per-message AI bill, no hallucinated rules.

The workflow, step by step

  1. Describe your idea. Tell your AI the premise, the level, the tone, and any beats you want.
  2. It drafts the story using the real building blocks — scenes, choices, skill checks, combat, loot, branching rules, and variables that remember what the player did.
  3. Validate. The AI asks ScryMaster “is this complete and playable?” — every choice leads somewhere, nothing is a dead end, and there’s a reachable ending. It fixes what’s flagged.
  4. Simulate. ScryMaster auto-plays the story a couple hundred times to confirm it can actually be finished and the difficulty feels right (how often players die, which endings get hit).
  5. You review & publish. Open it in the Studio, read it like a player, adjust anything, and publish it free to the store.

“How does it know the story actually works?”

Two automated checks, both available to your AI before anything is published:

  • Validate proves the structure is sound: a starting point, no broken links between scenes, no unreachable rooms, at least one ending you can actually reach, and only real commands.
  • Simulate proves it’s playable: out of (say) 200 automatic playthroughs, how many reached an ending, which scenes never got visited, and how deadly it is.

Being straight with you: today those checks cover structure and playability. They do not yet warn you about a scene with no narration or a missing picture — that polish-level linting is on our list. Until then, a quick read-through in the Studio (or asking your AI to “list any scene missing prose or art”) covers it.

“What if I edit it by hand afterward?”

Your edits and the AI’s edits are the same single story file. A well-behaved AI assistant re-reads the current version right before it changes anything, so it sees your manual tweaks and builds on them rather than clobbering them. And you can’t accidentally save a broken file: if a hand-edit leaves the JSON invalid (a missing comma, say), the Studio refuses to save it and shows you exactly where — your last good version stays safe on the server.

What you can build this way — and what you can’t (yet)

The AI path builds stories and everything inside them: scenes, dialogue, checks, encounters, loot, branching logic, variables, and the custom items/monsters an adventure ships with (they get bundled into the adventure so it’s self-contained).

It does not reach into your live game in progress — leveling up a character on the table, editing a party roster, or your campaign-wide homebrew library are managed in the app itself, not by the story-authoring AI. (An adventure can still grant a custom item or level the hero as part of the story — it just can’t reorganize your live campaign.)

Getting started today — the connector is live

  1. Mint a token. In the Studio, open any story → ⚙ Settings → Integrations → API tokens → create a token (it starts with scry_). Treat it like a password; you can revoke it any time.
  2. Connect the MCP server. Add ScryMaster to your assistant’s MCP config pointing at mcp/scrymaster.mjs (from the repo/package) with two settings: SCRYMASTER_TOKEN = your token, SCRYMASTER_API_URL = https://scrymaster.com.
  3. Author. Your assistant now has the full keyring: it grounds on the live vocabulary, creates or reads a draft, edits it, saves safely (a stale save is rejected instead of overwriting your work), validates, simulates, runs a dry publish for the full pre-flight verdict, and publishes — the same gates you’d hit in the Studio, and every AI edit is recorded in your story’s activity ledger so you always know who changed what.

No MCP client? The manual path still works: point your assistant at the live action vocabulary (and the OpenAPI spec), ask it to draft an adventure grounded in that list, then paste the result into the Studio’s JSON tab and hit Validate. Full endpoint details live in the API reference.

Next: The API & MCP reference →