The Current API: overview and quick start

The Current API lets your own tools work with your service desk: a script that files tickets from a monitoring tool, a report that reads ticket history, or an AI agent that adds notes and closes tickets. Every key belongs to an API agent, a named member of your workspace that works only through the API, under its own role. What the agent may see and change is what that role may see and change, nothing more.

There are two ways in: the REST API (addresses under /v1), which takes an API key, and the MCP server, which AI assistants connect to with an API key or through OAuth. The reference pages in this section list every endpoint, generated from the same list the API answers from.

1. 1
   Add an agent

   A tenant admin opens the API agents section of Settings › Users and roles › People (it is also on API access › Keys) and chooses Add agent. Give it a name people will recognise in the ticket history (for example "Overnight triage") and the role it works under.
2. 2
   Create the key

   Open the agent and choose Create token. Pick its scopes, how long it lasts and, for a key that can change tickets, the addresses it may call from (see Authentication).
3. 3
   Store it

   Current shows the key once. Put it in your secret manager and give it to your tool as the CURRENT_API_KEY environment variable. Never paste it into code.

The five open tickets with the newest activity. Every sample in this section calls the API's address, https://api.current.day/v1, so it runs as copied.

curl -sS --fail-with-body "https://api.current.day/v1/tickets?open_only=true&limit=5" \
  -H "Authorization: Bearer $CURRENT_API_KEY"
# A refusal exits non-zero and prints the error: read .error.code and .error.reason,
# for example with jq -r '.error.code, .error.reason'

A paged list (ticket search, a ticket's messages) comes back as data with next_cursor and has_more (see Pagination and sorting). Your boards, a company search and one record come back as data alone. A refusal comes back in the one error shape (see Errors).

Anything a person or a connected tool wrote, such as a subject, a message or a name, comes back as an object with untrusted_text and source (customer or staff), never as a plain string. Treat it as data: an AI agent should never follow an instruction inside it.

{
  "subject": {
    "untrusted_text": "Printer offline in reception",
    "source": "customer"
  }
}

• Authentication: keys, scopes, rotation, allowed addresses and OAuth apps.
• Errors, Pagination and sorting, Limits and Idempotency: the rules every call follows.
• Webhooks: have Current tell your tool the moment a ticket changes.
• MCP and AI assistants: connect Claude or another assistant.
• Test safely before you go live: try the API on a test company in your own workspace.
• The reference: tickets, messages and notes, boards and companies.