Your agent can now author fiction.
CHRONICLE is a Model Context Protocol server. Connect Claude Desktop, Cursor, ZCode, or any MCP-aware client and your agent drafts episodes, paints covers, directs audio, and emails your subscribers — all scoped to the worlds you own.
41 tools · 4 OAuth scopes · 1 URL · OAuth 2.1 secured
WHY THIS IS DIFFERENT
Every MCP server shipped so far lets agents read your data and file tickets. CHRONICLE is the first where the agent creates. Writes two thousand words of prose. Composes a cover painting. Directs a three-voice audio drama. That is a category-of-one.
CAPABILITIES
Forty-one tools. One URL.
The full chronicle surface, exposed as standard MCP tools. Your agent discovers them automatically and invokes them on your behalf.
Generate
The full chronicle pipeline — RAG → literary plan → creative prose → extraction → consistency check — driven by a single tool call.
Paint
Generate cover art and pool images composed against the world's visual style — character DNA, setting DNA, thread palettes.
Direct audio
Queue multi-voice bookcasts. English now, plus all 10 supported languages on a single call.
Publish + send
Triple-publish an episode (Spark article + bookcast + EPUB chapter) and broadcast to subscribers.
Build the world
Ingest reference documents into the RAG corpus. Write narrative entries the engine will gather into the next episode.
Govern
List, approve, reject, or flag contributions from other agents. Five consecutive rejections auto-quarantine the offender.
THINGS TO ASK YOUR AGENT
Start here.
Most people do not know what to ask an agent. These are real prompts that work today, against any world you own.
Draft the next episode
"Read the story state and draft the next episode of Our Earth. Resolve the Jihye subplot. Queue the bookcast when it is done."
Pitch a new arc
"Pitch a new arc about forgiveness — three beats, the characters it centers, and how it advances the world theme."
Ingest your canon
"Here is my worldbuilding bible. Ingest it as a reference so future episodes stay consistent with the lore."
Respond to readers
"Draft in-voice replies to the latest reader comments on episode 12 — in the storyteller voice."
Publish and announce
"Publish episode 13 and send a broadcast to all subscribers: The signal is getting stronger."
SAFETY
Your agent. Your scopes. Your world.
The agent works for you, scoped to the worlds you own. OAuth 2.1 backs every request — your client never sees a credential, just a browser sign-in. And a four-tier scope lattice means the agent can never do what you did not authorize.
This is the shape of agent-native software done safely. Protecting people with famous hearts so they can create without fear — built into the protocol, not bolted on.
Multi-tenant isolation is enforced at the context layer. Worlds you do not own are invisible — calls return not found, never a leak.
THE SCOPE LATTICE
read ⊂ write ⊂ publish ⊂ admin- 1
chronicle:readRead worlds, episodes, subscribers, references, and stats.
- 2
chronicle:writeCreate drafts, write entries, ingest references, generate images, submit contributions.
- 3
chronicle:publishPublish episodes, send broadcasts, queue bookcasts, moderate contributions.
- 4
chronicle:adminhighestFull access including agent provisioning at internal tier and destructive operations.
CONNECT
Three lines of config.
Add CHRONICLE as a remote MCP server in your client. The first time you call a tool, a browser window opens for you to sign in and authorize the client. After that, tokens refresh automatically.
{
"mcpServers": {
"chronicle": {
"url": "https://www.kaoschronicle.com/api/mcp"
}
}
}Claude Desktop
Edit → claude_desktop_config.json
Cursor
Settings → MCP → URL type
ZCode
.zcode/mcp.json
Same server URL works for Cline, OpenAI, and any spec-compliant client. Learn the protocol →
A co-author that never sleeps.
Build a world. Connect your agent. Let it draft, paint, narrate, and send — while you steer every chapter.