
Jira Service Management AI approvals in your queue
Argent Strategic Intelligence opens a Jira Service Management ticket for AI budget and MCP access requests, writes the decision back, and applies a native Approve or Decline under its own checks.
Approvals that land in your ticket queue
Two kinds of request in Argent wait for a person's yes: an owner asking for more AI budget, and an admin or operator asking for an MCP server or tool. Connect Jira Service Management and each request also becomes a ticket your service desk already works from.
- Argent opens the ticket, reads it back, and writes the approve or deny decision onto it.
- Every decision runs through Argent's own checks and never waits on Jira Service Management. If Jira Service Management is slow or down, the decision still commits and the ticket update is retried.
- One connection serves both request types: the AI spend caps approvals and the MCP Gate requests behind agent reach control.
Your change and access process stays where your auditors already look. Argent adds the AI requests to it.
What Argent does with Jira Service Management
| Step | What happens |
|---|---|
| Open | A ticket for every new request, when you switch that on, or one by hand from a request with CREATE TICKET. Each request gets at most one live ticket per ticketing system. |
| No duplicates | Every ticket carries Argent's own request id as a label, argent-<id>. Before creating, Argent looks that id up, so a retry after a timeout adopts the ticket that already exists instead of filing a second. |
| Read back | Argent reads the ticket's state and assignment group back on a schedule, and on SYNC NOW. Closed tickets stop being read. |
| Decide | The approver decides in Argent, or on the ticket itself (below). The decision commits first; the update to the ticket is queued and retried until it lands. |
| Record | Every connector and ticket change is a signed audit row. |
Approve or decline right on the Jira ticket
For budget requests, an approver can also use Jira Service Management's own Approve and Decline buttons, and Argent applies that answer.
- Filed as a customer request. When you set a request type, Argent files the ticket as that request type, so the approval steps you set up in Jira Service Management attach to it.
- Read on Argent's own poll. Argent reads the settled approval back through the same guarded connection. There is no new inbound route into Argent.
- Matched to a real admin. Argent matches the Jira approver to an active Argent administrator of this tenant, by the account mapping you set on the connection or, for an account you have not mapped, by the email Jira reports. The decision is then applied as that administrator, through the console's own decision path: the same fences, the same funding checks, the same signed decision event.
- Any doubt refuses. An approver Argent cannot match to an active administrator of this tenant, a role too low to approve, ticket text edited after Argent wrote it, or a requester approving their own request changes no budget. Argent signs the refusal and comments on the ticket.
The service desk keeps its buttons. Argent keeps the rules.
How it connects
Argent calls Jira Service Management over HTTPS on the standard port and nothing else.
- Narrow egress. Only
httpsURLs, and only a host under*.atlassian.netor one exact host your administrator adds. Plain http, another port, a private address or an off-host redirect is refused before any request leaves. - No secret in Argent's database. The Jira Service Management credential stays in your vault; Argent stores only a reference to it, reads it when it connects and holds it in memory. It never appears in configuration, a status, a log or an error. Works with HashiCorp Vault and OpenBao, AWS Secrets Manager, CyberArk Conjur or Infisical.
- Plain errors. A failed test reports a fixed reason (blocked by egress policy, DNS lookup failed, TLS verification failed, unreachable, or the HTTP status) and never repeats text from Jira Service Management's response.
- Tenant fenced. Each tenant has its own connection; tenants never see each other's tickets or settings.
What you configure
A SuperAdmin or TenantAdmin connects Jira Service Management once for the tenant.
- Store the credential for a Jira Service Management account that may create and update tickets in your vault, and copy its reference.
- Add the connection in the Argent console under Integrations, Notifications, Ticketing: the Jira Service Management address, where tickets go (the project, and optionally a request type), and the vault reference. Optionally map Jira approver accounts to the Argent administrators they act as (an unmapped approver is matched by the email Jira reports).
- Test, then connect. Test makes a real read-only call to Jira Service Management. Connect runs the same check and only records the connection as connected if it passes.
- Choose what opens a ticket. On the Budgets approvals page, Open a ticket for every new request; the same automatic setting exists for MCP requests. Leave either off and tickets are opened by hand.
- Or script it.
argentctl integrations applyreads one JSON document, stores the secret in the tenant vault, then creates, connects and tests the connection, and never prints a secret.
Org and department admins see which ticketing system is connected and its state; only SuperAdmins and TenantAdmins see the host and the credential reference.
Every step is on the signed record
Configuring, testing, connecting, disconnecting and every ticket write leave a signed row in Argent's audit ledger.
- A configuration change records whether the host, port or credential reference changed, as flags, never the values.
- A request's reason reaches the ledger only as its length and a keyed fingerprint, so free text never sits in the signed record.
- If the audit ledger cannot be written, the change is refused and nothing is saved.
Hand the record to an auditor with one-click audit evidence. See audit and accountability.
Where this fits
- Capabilities: AI spend caps (budget requests and approvals), agent reach control (MCP server and tool requests), keyless agents (credentials stay in the vault).
- Goals: control budget and token spend, audit and accountability.
- Use case: controlling agent tool access.
- Also in ticketing and approvals: ServiceNow, and the work trackers Jira, GitHub, GitLab, Linear and Notion.
- Alerts for the same requests: Slack or email notifications.
Browse all integrations.
Want this running on your own use cases?
Book a demo