Plan Tiers and Tool Gating

Every StackJack tool has a required plan tier, and every connector you subscribe to has a plan. This page explains how the two interact, what each tier unlocks conceptually, and what happens when a call hits a limit.

Plans are per connector, not per account

StackJack subscriptions are per connector. Your tenant might run HaloPSA on Pro and NinjaRMM on Free at the same time. When the AI calls a halo_* tool, StackJack checks your Halo subscription's plan; when it calls a ninja_* tool, it checks your NinjaRMM subscription. There is no single account-wide tier.

The available plans are Free, Pro, Business, and Enterprise. Current pricing for each is shown in the portal and at checkout.

What each tier unlocks

Every tool is tagged with the minimum plan needed to call it. You can see each tool's tier badge in a connector's API Permissions planner and in the tool selector on Endpoints. Conceptually:

• Free — the core read surface of each connector: listing and fetching tickets, devices, companies, assets, and other day-to-day lookups. Several connectors (for example Pax8, Kaseya Quote Manager, ConnectWise Sell, and Telivy) place their entire tool set in Free — for those, paid tiers differ only in the monthly call allowance.
• Pro — write operations (create, update, delete), actions (run scripts, trigger jobs), and advanced or analytical tools, on top of everything in Free.
• Business — the same tool set as Pro on every connector, with a much higher monthly call allowance. No connector has Business-only tools. Business is the "same tools, more volume" tier.
• Enterprise — all plan-tiered tools, with no monthly call cap. Permissions, connector configuration, feature flags, and the other runtime gates still apply.

Exact per-tier tool listings for each connector are published in the generated connector tool catalogs.

Monthly call allowances

Each connector subscription includes a monthly tool-call allowance, enforced per billing cycle:

Details that matter in practice:

• Only successful live connector calls count. A call StackJack records as failed — whether blocked by a gate or failed at the vendor API — does not consume your allowance. A successful replay served from a recorded fixture also does not consume live connector quota.
• The cycle is your billing cycle, not the calendar month. The counter resets when your connector subscription renews. Your dashboard's per-connector usage bars track the same cycle the enforcement uses.
• StackJack can set a custom limit on a subscription — if you have arranged a custom allowance, the enforced number is the one shown on your dashboard.

TechTribe call-volume perk

If your organization redeemed the StackJack TechTribe offer, the enforced monthly allowance has a permanent tenant-wide floor of 7,500 calls on Pro and 75,000 calls on Business. Free and Enterprise are unchanged. The floor applies to existing and future connector subscriptions, and the portal's usage meters show the effective limit. It never replaces a higher custom limit or reduces an allowance an operator has set above the perk floor.

There is no general connector-call burst limit

StackJack does not impose a plan-based per-minute or burst limit on ordinary connector-tool traffic. The per-connector billing-cycle allowance is the product-plan enforcement point. Responses include X-RateLimit-*-Burst headers. They are informational only and do not describe a traffic budget, so do not build tooling against them.

This is not a promise that every platform workflow is unbounded. Support-ticket creation has separate anti-abuse limits, Automation runs are subject to capacity and concurrency controls, public catalog endpoints have their own protection, and connector clients pace outbound vendor traffic. Those controls are independent of the connector plan allowance described here.

Two things can still slow down or reject individual calls:

• Vendor API protection. Connector-specific pacing, concurrency controls, and retry policies reduce the risk of overrunning a vendor's published limits. They cannot guarantee that a vendor will never throttle or ban a key: cancellation, a time budget, a local concurrency wait, or a vendor response can still end or refuse a call.
• Vendor rate limits. If the vendor still returns a "too many requests" response, the tool call returns a structured rate_limited error and the AI is told to wait and retry. See Tool errors and troubleshooting.

Runtime gates: what can be checked before a connector call

Platform tools and connector tools do not all take one identical path. A connector call passes the gates that apply to its client and run, and a refusal returns a specific structured error rather than failing silently. The common order is:

1. Client tool selection — is the tool enabled for this MCP client and team member? If not: tool_not_allowed.
2. Automation safety, when applicable — dry-run write blocking and destructive-consent checks run before a live connector action.
3. Credential reachability — is the subscribed connector configured, and did its credentials load for this request? If not: connector_not_configured or the retryable connector_unavailable.
4. Plan and subscription health — is the required tier in plan, is payment active, and is the billing period current? Possible codes include plan_upgrade_required, subscription_past_due, subscription_paused, subscription_canceled, subscription_inactive, and subscription_period_ended.
5. Monthly allowance — is the cycle counter under the effective limit? If not: monthly_limit_exceeded (the error includes usage, limit, and the billing-cycle window).
6. Instance version, when declared — if a connected instance reports a version below the tool's floor: instance_version_unsupported.
7. Run governance, when applicable — ordered destructive-chain, replay-fixture, approval, and result-size checks can add further Automation-specific refusals before or after the vendor call.

Each error includes plain-language guidance the AI can relay to you. The full error reference is in Tool errors and troubleshooting.

What a downgrade or cancellation looks like

When a connector subscription is downgraded or canceled:

• Tools above the new plan tier (or all of the connector's tools, on cancellation) disappear from the harness's tool list the next time it refreshes its tools.
• Saved tool selections that reference now-unavailable tools behave differently by surface: in the automation tool picker they are flagged as stale with a one-click repair; in the MCP-client and team-member editors they are simply hidden from the checklist and excluded from the selection count (details).
• If the AI calls a no-longer-available tool before its list refreshes, it receives the specific gate error rather than a generic failure.

Checking your own gating live

Ask your connected AI to run:

• stackjack_session_info — shows each active connector, its plan, its monthly call limit, and how many tools are accessible to this client versus the total.
• stackjack_list_tools — lists platform tools plus tools for connector credentials that loaded on this request, with their required plan and an accessible: true/false flag. A transient credential-load failure can omit a subscribed connector from this diagnostic projection even while protocol discovery keeps that configured connector's direct tools visible; use stackjack_health_check when a family is unexpectedly absent.