What is the WorldOS MCP server?
WorldOS MCP is a remote Model Context Protocol server that connects your AI agent to the WorldOS world-building platform. Instead of copying generated text into an editor, your agent can assemble a playable world, configure its apps, write characters, and edit maps through tools connected to your account. You describe the setting and the experience you want; the agent reads the authoring guidance and builds the world with persistent app data.
The WorldOS MCP server supports both new projects and changes to worlds you already own. An agent can discover existing apps, validate its work, inspect the current world, and make focused edits. Newly created worlds start unlisted so you can review them before publishing. Changes to an already-public world update its live content, so tell your agent exactly what you want to change.
What it can do
The same building blocks WorldOS creators use, shaped into safe tools an agent can combine.
Worlds
Author, validate, patch, and explicitly publish worlds you own.
Maps
Regions, factions, ownership, labels, and region stats — read, validate, patch.
Playtests
Isolated temporary turns against an owned world; no real save is touched.
Apps
Search the catalog, then create and update localized interactive widgets.
Characters
Author library characters and cast them into worlds you own.
Saves
Create real saves, advance billed turns, and page through turn records.
Explore the tool reference
Guidance & discovery
get_authoring_guidesearch_appsget_app_guide
World authoring
inspect_world_payloadvalidate_worldcreate_world
Owned worlds
get_owned_worldlist_owned_worldsupdate_worldpatch_worldcreate_world_cover_uploadcomplete_world_cover_uploadcreate_world_asset_uploadcomplete_world_asset_uploadvalidate_world_for_publishpublish_worldget_world_summary
Maps
validate_mapget_world_mappreview_world_map_replacementset_world_mappatch_world_mapget_world_map_backupsrestore_world_map_backup
Isolated playtests
start_world_playtestget_world_playtestget_world_playtest_historyplaytest_world_turndelete_world_playtest
Apps
validate_app_draftcreate_app_draftget_owned_applist_owned_appsupdate_app_draft
Characters
create_characterget_owned_characterlist_owned_charactersupdate_charactercreate_world_characterupdate_world_character
Real saves
create_owned_saveplay_owned_save_turnlist_owned_savesget_owned_saveget_owned_save_turns
Connect your client
Every client uses the same endpoint. Sign-in happens in your browser through WorldOS OAuth — you never paste an access token.
- 1. Choose an MCP client that supports remote Streamable HTTP and OAuth. Add the WorldOS endpoint shown above; no local WorldOS server installation is needed.
- 2. Sign in to WorldOS in the authorization window, review the requested permissions, and approve the connection for your own account.
- 3. Ask your agent to read the WorldOS authoring guide and list your owned worlds. Confirm it can read your account before asking it to create or edit anything.
claude.ai on the web — add a custom connector, then enable it inside the chat.
- 1
Add the connector
Open Settings → Connectors, click Add custom connector, name it WorldOS and pastehttps://worldos.cc/api/mcp. - 2
Confirm and add
Leave Advanced settings empty (no OAuth Client ID/Secret needed), accept the trust confirmation, then click Add. - 3
Connect with OAuth
On the WorldOS card in your connector list, click Connect — a WorldOS window opens to sign in and approve the access. - 4
Enable it in your chat
In the conversation, open the + (search & tools) menu near the message box and toggle WorldOS on — connectors are per-conversation, and individual tools can be toggled under Search and tools. - 5
Ask for something
Try "Create a cyberpunk detective world on my WorldOS account." The free plan allows one custom connector; connectors sync with the Claude desktop app.
Build your first world with WorldOS MCP
Once connected, describe a concrete setting, the player’s role, and the apps you need. You can ask the agent to inspect and validate the result before you open it in WorldOS.
Use WorldOS MCP to create an unlisted mystery world set aboard a research ship. The player is the captain. Add three crew members, private chats, a story window, and an inventory with the ship’s log. Give each character a clear motive and write an opening scene. Reuse existing apps, validate the world, and return its link for my review. Do not publish it.
The agent should create a world owned by your account and return a link you can open. Review the opening, characters, and app state, then request specific changes. Publishing is a separate action that you explicitly request when the world is ready.
WorldOS MCP questions
- Do I need a WorldOS account or an API key?
- You need a WorldOS account and a compatible MCP client. WorldOS uses OAuth 2.1 to authorize the connection, so you do not paste a WorldOS password or access token into the client’s server configuration. Your AI client may have its own subscription or model charges.
- Can my agent test or play a world?
- Yes. Isolated creator playtests use temporary sessions and do not create normal player saves or contribute to play statistics. Real-save play is a separate capability that requires your consent, uses normal save slots, and may spend Zaps when advancing turns. Tell the agent which kind of test you intend.
- What if my MCP client cannot connect?
- Check that the endpoint is exactly https://worldos.cc/api/mcp and that the client supports remote Streamable HTTP with OAuth. Complete the browser authorization flow and enable the connector in your conversation if the client requires it. If you still need help, contact WorldOS support with the client name and the error message; do not include passwords or tokens.
- Will the agent publish automatically?
- A newly created world starts unlisted. Publishing requires your explicit request, successful validation, and ownership by the authorized account. Editing a world that is already public updates its live content; ask the agent to inspect the current version and explain the intended changes before editing.
The agent edits. You choose when to publish.
Your account owns the result — everything the agent creates belongs to the account that approved the OAuth connection.
Explicit publishing — publishing a new world requires your explicit request, successful validation, and ownership by the authorized account.
Conflict-safe editing — updates require the latest version timestamp, so agents never silently overwrite each other.
Validated and auditable — payload limits, content checks, ownership checks, and tool-call audit logs run server-side.
