Connect Neon.

Which key to create

The Neon connector uses the Neon API at https://console.neon.tech/api/v2 with a bearer token, per the API reference. Neon's API keys page describes 3 kinds of key. Personal keys are created by any user under Settings → API keys and reach every organization project the user is a member of. Organization keys are admin-level and reach everything. Project-scoped keys are created by an organization admin, cover a single project, and carry Editor access, which the page describes as able to “read and modify project resources but cannot delete the project or manage who can access it.”

Use a project-scoped key for the project you want watched. It is the narrowest key Neon offers, and it is enough for both the reads and the kill. Create 1 key per project; a key per project is what makes the audit log say which database a kill touched.

Create the key

1. As an organization admin, open the organization's Settings → API keys page in the Neon Console.
2. Create a project-scoped key, pick the project, and name it mepa8 followed by the project name.
3. Copy it once. Neon shows it at creation; Mepa8 shows the last 4 characters after save and never the rest.
4. In Mepa8, Add vendor → Neon → paste → Save. The first check runs at once.

What Mepa8 reads

Three calls per check. The consumption endpoint GET /consumption_history/v2/projects returns compute seconds and storage for the project over a window, which the connector turns into a month-to-date figure and an on-pace figure at your plan's rate. The project's endpoints list gives each compute's state, its autoscaling limits and its suspend timeout, which is how Mepa8 can tell you the compute never sleeps. The project record gives the plan. Rates come from the pricing page: $0.106 per CU-hour on Launch, $0.222 on Scale, $0.35 per GB-month of storage, read October 9, 2026.

Our own database, before the fix on October 8, 2026, averaged 0.64 CU around the clock on the Scale plan, which the connector reports as $102 on pace for a 30-day month before storage. After the fix the churn on the hottest table went from 251 updates in 60 seconds to 0, and the pace is expected to settle near the 0.25 CU floor, about $40. The incident write-up has the whole trace.

Alerts this connector raises

• Warning when the on-pace figure passes the budget, critical at 150% of it.
• Warning when a compute has been active for 24 hours without a suspend on a plan where scale-to-zero is enabled.
• Warning when autoscaling's maximum changed since the last check, because a raised ceiling is where runaway loops hide.
• Critical when the API returns a project in a non-active state or a payment problem.

“Scale to zero: after 5 min, can be disabled.”Neon pricing page, Launch plan, read 2026-10-09

The kill, by API

Neon is Full tier because the kill is an API call with an API undo. Confirming a kill sends POST /projects/{project_id}/endpoints/{endpoint_id}/suspend, and restore sends the matching /start; both are listed in the API reference alongside a /restart. A suspended compute bills nothing and answers nothing, so every application reading that database stops at once. The propose step says so, with the project name and the number of endpoints it will suspend, before the token is issued. Nothing is deleted: branches, data and roles are untouched, and start brings the compute back in the state it was in.

Questions people ask

Straight answers

Can Mepa8 read my data with this key?

The project-scoped key carries Editor access to project resources per Neon's API keys page, which includes connection details. Mepa8 only calls the consumption, project and endpoint routes, logs every call in the audit log, and stores the key encrypted. If you want a stricter boundary, run the monitor with a personal key from a Viewer-role user.

What exactly happens when I confirm a kill?

One POST to the suspend route for each compute endpoint you selected, listed in the propose screen by name. The database stays intact. Restore sends the start route. Both appear in the audit log with your name, the time and the endpoint IDs.

How often does Mepa8 call the Neon API?

Three calls per check, 24 checks a day on the hourly schedule, so 72 requests a day per project. Consumption data in Neon's API lags real time, so the on-pace figure is a trailing number and the first reading appears after the second check.

Sources

1. Neon — manage API keys — read 2026-10-09
2. Neon — API reference (consumption, endpoints suspend/start) — read 2026-10-09
3. Neon — pricing — read 2026-10-09

Change log: October 9, 2026 — First published. Key types and endpoints read from the Neon docs the same day.