Errors & limits

A failed tool call returns an error the agent can read, in words that say what went wrong and what to do next. Every error on this page means nothing was changed, unless the error says otherwise.

Failure	Message	What to do
Invalid arguments	”Invalid arguments for get_event:” then one line per field, for example “Unknown argument “eventId”. Valid arguments: event_id.”	The call never ran. Fix the named fields and call the tool again. A misspelled or extra argument is refused, never ignored.
Not authenticated	An HTTP `401` before any tool runs, with a `WWW-Authenticate` challenge.	Your OAuth session expired or was revoked. Reconnect. See Connecting.
No organization access	”You don’t have access to any organization’s events.”	Your account is not in an organization with events. Ask your Nowadays contact.
Organization required	”You belong to multiple organizations. Call select_organization with the id of one of: …”	The message lists each organization’s name and id. Call `select_organization` with one id, then retry.
Not found	”Event not found or not accessible to your organization.”	The event does not exist, or it is outside what you can read. Check the id, or call `find_event`.
Timed out	”That took too long, so nothing came back and nothing was prepared or changed. Try again.”	Returned by a read or a prepare. Retry the call.
Timed out, may have saved	”That took too long to answer, so it may or may not have saved. Read it back before trying again.”	Returned by `shortlist_proposal`, which saves at once. Read the event back before you retry.
Rejected write	The reason, for example “This event is canceled and cannot be edited. Reopen it in Nowadays first.”	Correct the call and retry. See Making changes.
Refused confirm	The reason, for example “This approval expired, so nothing was changed.”	Prepare the change again and show the new preview. See `confirm_change`.
Unexpected error	”Something went wrong on our side. If this call changes something, read it back before you retry. If it fails again, contact Nowadays support.”	Something failed on our side. A read changed nothing. After a write, read it back first, then retry once.

Not found covers both cases

A tool reads an event only if your own access admits it and the organization you selected was granted it. An event outside that scope returns the same not-found error as one that does not exist. This is on purpose, so you never learn whether another organization’s event exists. Don’t treat the error as proof that an id is invalid. It means only that you cannot read the event. A tool returning fewer events than you expect is usually access, not a bug. Enterprise events are private to their creator and their approvers by default.

Refused confirms

confirm_change refuses in these cases. Each message is exact.

Case	Message
Expired	”This approval expired, so nothing was changed. Prepare the change again, show the person the new preview, and wait for their yes.”
Not approved	”The person must approve this change first, so nothing was changed. Show them the preview and ask. Only if they say yes, confirm the same approval_token and change again with user_approved: true.”
Different change	”The change in this call is not the one that was prepared, so nothing was changed. Show the person the preview, then confirm the same approval_token again with change copied exactly from the prepare result.”
Someone else’s	”This change was prepared by someone else, so only they can confirm it. Nothing was changed.”
Invalid token	”This approval is not valid, so nothing was changed. Prepare the change again.”
Organization changed	”Your selected organization changed since this was prepared, so nothing was changed. Prepare it again.”
Event out of reach	”You can no longer reach this event in the organization you have selected, so nothing was changed. Prepare it again if you still can.”
Event changed	”This event changed since the preview, so nothing was changed. Call get_event, then prepare the change again.”
Record unreachable	”Could not reach the record of this change, so its state is unknown. Call confirm_change again with the same approval_token and change before preparing anything.”

A refused write or confirm means nothing was written, so it is always safe to correct the call and retry without first checking what partially landed. The one exception is “Record unreachable”. Confirm again with the same token, which never runs the change twice.

Every call is recorded

Nowadays records every tool call: who made it, from which MCP client, in which organization, with which arguments, and how it ended, including refused and invalid calls. An approval token is never recorded.

Time limits

Each tool that works on sourcing, venues, or proposals has its own time limit, from 5 to 30 seconds. A read or a prepare that runs longer returns the timed-out error, and nothing was prepared or changed. Retry it once. If it times out again, narrow the call, for example with fewer cities or a venue filter. shortlist_proposal saves at once, so a save can still land after its time limit. It returns the “may have saved” error instead. Read the event back with get_proposals before you call it again.

Rate limits

Requests are rate limited per user, across every MCP client you have connected. Limits are approximate and enforced on a best-effort basis. Exceeding the limit returns a rate-limit error. Back off and retry, increase the delay on repeated failures, and spread bulk work out rather than firing many calls at once. If you have a legitimate need for a higher limit, contact us. We can raise limits on request.

Getting started

Tools

Reference

Not found covers both cases

Refused confirms

Every call is recorded

Time limits

Rate limits

​Not found covers both cases

​Refused confirms

​Every call is recorded

​Time limits

​Rate limits

Not found covers both cases

Refused confirms

Every call is recorded

Time limits

Rate limits