Continuity by Ouroboros Apps
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Continuity by Ouroboros Apps check my story bible for Kael's eye color before writing the next scene"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Continuity
Continuity keeps your project facts across chats. It's a story bible for fiction writers: your characters, places, timelines and plot facts stay consistent while you write in ChatGPT, Claude, Grok or Cursor.
It also works with Gemini and any other MCP client that supports Streamable HTTP and OAuth. Made by Ouroboros (https://ouroborosapps.com).
MCP address:
https://continuitywriter.com/mcp
Sign in with your Continuity account when the assistant opens OAuth. Do not paste an API key or password into a header. Story tools need an active Continuity account; see https://continuitywriter.com for details.
What the assistant can do
After you approve the connection, the server exposes these tools:
list_story_projects
create_story_project
get_continuity_context
search_story_canon
upsert_canon_entry
get_canon_history
list_approved_scenes
record_approved_scene
save_story_checkpoint
list_story_checkpoints
continuity_audit
Locked facts stay locked until you revise them. Checkpoints are provisional and are not canon. The assistant only calls these tools when you and the host allow it.
Related MCP server: storyai
Permissions, data, and limits
OAuth scope. Sign-in is OAuth 2.1 with dynamic client registration and PKCE. The scopes are email and offline_access (refresh tokens). A token reaches only the signed-in user's own story projects; database ownership rules keep one account from reading another's projects.
Read operations. list_story_projects, get_continuity_context, search_story_canon, get_canon_history, list_approved_scenes, list_story_checkpoints, and continuity_audit only read (readOnlyHint: true).
Write operations. create_story_project creates a project. save_story_checkpoint saves a provisional session note. upsert_canon_entry creates or revises a canon entry, and record_approved_scene saves an approved scene summary. The last two are marked destructiveHint: true because they can replace saved text.
Delete operations. No MCP tool deletes anything. Projects are exported or deleted by the user in the Continuity workspace at https://continuitywriter.com.
Confirmations and safeguards. Tool descriptions tell the assistant to save canon and scenes only after the user approves them; brainstorming is not approval. The destructive hints let hosts ask before running those two tools. A canon revision must send the expectedRevision it read, so a stale edit fails instead of overwriting, and every revision is kept in get_canon_history. An occupied scene number is kept unless replaceExisting is true and the user asked for the replacement. Identical retries of the canon, scene, and checkpoint tools do not create duplicates.
Data stored. Account id and sign-in email, subscription status and Stripe references, and the story content the user saves: projects, canon entries and their revisions, approved scene summaries, and provisional checkpoints. Tools receive only their own arguments; Continuity does not receive whole conversations. Supabase hosts authentication and the database, Vercel hosts the server, and Stripe handles payments (Continuity never stores full card numbers).
Retention and privacy. Saved stories stay until the user deletes the project or asks for account deletion. Deleting a project removes its live canon, revisions, scenes, and checkpoints; provider backups follow the provider's retention schedule. Story content is not sold and is not used to train models. Full policy: https://continuitywriter.com/privacy
Free limits. Tool discovery (initialize, tools/list) is open. Story tools need a Continuity account with a 14-day trial, then a monthly or yearly subscription through Stripe Checkout. Stripe shows the amount; this repository does not list one. Text fields are capped (canon content 12,000 characters, checkpoint state 6,000) and list tools page 50 rows at a time.
How it is used. A writer creates a project, approves character, place, and timeline facts as canon, and records each approved scene. In a new chat, the assistant calls get_continuity_context to pick up where the story left off, and runs continuity_audit on a draft scene to catch contradictions with locked canon before the writer accepts it.
Connect
Cursor, in ~/.cursor/mcp.json or a project .cursor/mcp.json:
{
"mcpServers": {
"continuity": {
"url": "https://continuitywriter.com/mcp"
}
}
}Claude Code:
claude mcp add --transport http continuity https://continuitywriter.com/mcpOther clients: add the same URL, choose OAuth, and leave client id and secret empty. Continuity supports dynamic client registration. Full steps for each assistant are on the connect page.
Grok Build: install the ouroboros-continuity plugin from the plugin marketplace, then sign in when the browser opens.
Registry metadata for this remote server is in server.json (io.github.LAHutchins91/continuity).
Plugin contents
This repo is a plugin for Grok Build and Cursor, published by Ouroboros Apps:
Grok Build:
.grok-plugin/plugin.jsonplus.mcp.jsonCursor:
.cursor-plugin/plugin.jsonplusmcp.jsonOne skill:
skills/story-bible/SKILL.md(instructions only, no scripts)One remote MCP server,
continuity, over Streamable HTTP
The plugin has no hooks, commands, agents, or LSP servers, and it runs no code on your machine. The src/ folder and Dockerfile are MCP server source, published for reference and registry container checks. When someone runs that code as a server, it reads its own server settings (database and billing keys) from that server's environment. The plugin does not build, install, or run it, and nothing in this repo runs on install (no postinstall or other lifecycle scripts).
Network endpoints and credentials
https://continuitywriter.com/mcpis the only MCP endpoint. Story tools read and write the story projects in your own Continuity account.Sign-in uses OAuth 2.1 with dynamic client registration. Protected resource metadata is at
https://continuitywriter.com/.well-known/oauth-protected-resource. The authorization server it lists is Continuity's account service, hosted on Supabase (https://kufinkiktgnnhusuenjq.supabase.co/auth/v1). Your browser opens it when you sign in.Credentials: your Continuity account, through that browser sign-in. The plugin needs no API key, environment variables, or local files, and it does not read any.
Requests without a valid token get
401 Unauthorized.Privacy: https://continuitywriter.com/privacy. Support: https://continuitywriter.com/support. Contact: ouroborosplugins@gmail.com.
More from Ouroboros: https://ouroborosapps.com
Available Tools
11 toolscontinuity_auditcontinuity auditBRead-onlyIdempotentInspect
Retrieve locked canon for the assistant to compare against a short proposed scene. This tool supplies evidence; the assistant must identify contradictions. Does not save or approve the scene.
Inputs: projectId (required), proposedText (required), offset. Returns: data
create_story_projectcreate story projectBInspect
Create a new private story project when the user asks. Does not save canon or scenes.
Inputs: name (required), description. Returns: data
get_canon_historyget canon historyBRead-onlyIdempotentInspect
Read previous and new content for a saved canon entry, newest first. No changes are made.
Inputs: canonEntryId (required), offset. Returns: data
get_continuity_contextget continuity contextARead-onlyIdempotentInspect
Retrieve relevant canon, recent approved scenes, and the latest provisional checkpoint before continuing a saved story or recovering it in a new chat. request is a brief scene/topic query, never a chat transcript. Results are a selection; use search_story_canon for specific missing facts.
Inputs: request (required), projectId (required), limit. Returns: data
list_approved_sceneslist approved scenesARead-onlyIdempotentInspect
Read approved scene summaries in story order. Page with offset to inspect older scenes or verify a scene number before saving.
Inputs: projectId (required), offset. Returns: data
list_story_checkpointslist story checkpointsARead-onlyIdempotentInspect
Retrieve provisional story handoffs to recover progress across chats. These notes never override approved canon. Page with offset for older checkpoints.
Inputs: projectId (required), offset. Returns: data
list_story_projectslist story projectsARead-onlyIdempotentInspect
Find the user’s saved story projects before starting or resuming a story. Use the returned ID; do not guess projects. Page with offset.
Inputs: offset. Returns: data
record_approved_scenerecord approved sceneADestructiveIdempotentInspect
Store a concise summary of a scene the user has approved. Never record an unapproved draft. An occupied scene number is preserved unless replaceExisting is true and the user requested replacement. Identical retries return the saved scene.
Inputs: summary (required), projectId (required), sceneNumber (required), title, +2 more. Returns: data
save_story_checkpointsave story checkpointAIdempotentInspect
Preserve a concise story-session handoff when the user asks to save progress or has authorized ongoing checkpoints for this story. Stores provisional notes, not canon or an approved scene. Separate established progress from unresolved ideas. Send only focused story state, never a full transcript, credentials, or unrelated personal details. Reuse checkpointId on retries.
Inputs: title (required), projectId (required), storyState (required), checkpointId (required), +2 more. Returns: data
search_story_canonsearch story canonARead-onlyIdempotentInspect
Search saved canon by literal title/content text, or browse every entry with an empty query and offset. Use to verify details before revising an entry or when context retrieval omits something.
Inputs: projectId (required), query, offset. Returns: data
upsert_canon_entryupsert canon entryADestructiveIdempotentInspect
Save a fact or revise canon only after the user approves it. Preserve the existing title/kind. LOCKED requires approval as established fact; brainstorming is not approval. Existing entries require expectedRevision from retrieval; conflicting edits fail without overwriting. Identical retries leave revision history unchanged.
Inputs: kind (required), title (required), status (required), content (required), +4 more. Returns: data
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 11 tools
Most tools map cleanly to a resource+action (scenes, projects, checkpoints, canon writes), so selection is straightforward. The one soft cluster is the three read paths over canon — get_continuity_context, search_story_canon, and continuity_audit — which all return canon-ish content, though the descriptions draw usable lines (bundle vs. literal lookup vs. locked-canon evidence for contradiction checks).
Ten of eleven tools follow a clear verb_noun pattern (get_, search_, upsert_, list_, record_, create_, save_). The lone outlier is continuity_audit, a bare noun_verb with no leading action verb, and the domain prefix drifts between story_ and canon_ across otherwise parallel tools.
Eleven tools is well within the sweet spot and each one covers a distinct surface: context recovery, canon read/write/history, scenes, projects, and checkpoints. No tool looks redundant or filler.
Core lifecycle is covered: canon has upsert plus literal search and revision history, scenes have list and record (with replaceExisting), checkpoints have save and list, projects have list and create. The notable gap is that nothing is ever deletable — no canon removal, no project rename/delete, no checkpoint pruning — which is plausibly intentional for an append-only continuity log but leaves dead ends for cleanup.
Maintenance
Related MCP Connectors
One memory, every AI. A shared, user-owned markdown memory your AI clients read and write over MCP.
Persistent memory for AI assistants. Save once; recall from Claude, ChatGPT, or any MCP client.
Persistent memory for AI assistants — one shared, OAuth-secured vault for every MCP client.
AI second brain for people who coordinate several projects. Search and write to your daily notes, tasks and boards, calendar, meeting recaps, documents, clips and an Atlas of people, organizations and projects from Claude, ChatGPT or any MCP client. Free plan, OAuth sign-in.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceMCP server for managing a writer's bible, a structured and searchable knowledge base of a narrative universe with tools for characters, places, events, and semantic search.-
- AlicenseBqualityDmaintenanceEnables writers and AI agents to preserve continuity in long-form fiction by maintaining a narrative knowledge graph and exposing MCP tools for querying outlines, entities, references, and consistency diagnostics.14MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI-driven long-form novel creation and management, including chapter generation, character and timeline tracking, semantic memory retrieval, version savepoints, and deep consistency checking across multiple projects through MCP tools, with support for local LM Studio or any OpenAI-compatible API.16MIT
- AlicenseAqualityCmaintenanceProvides a persistent, branching store for fictional scenes — stories, character assignments, turns and editable summaries — exposed as ten validated, revision-checked MCP tools. Supports optimistic concurrency, idempotent turn appends, canon rules and story export over a bearer-token-authorized HTTP boundary that fails closed.10MIT