Terminal
Terminal/Skill
Models Konnect usage based billing as meters, features, plans and rate cards, then checks allowance and overage math.
By Kong
Works with
npx skills add kong/ai-marketplace --skill konnect-monetizationTurn a stated pricing model into a correct Konnect billing catalog — meters, features, plans, rate cards — and prove the money is right before a customer is billed against it.
Own the modeling decision and the validation loop. A pricing model is almost always under-determined: it names prices and allowances and stays silent on what one billable unit is, what happens past the allowance, and whether the base fee survives zero usage. Resolve those silences with the user, not alone.
Do not absorb AI Gateway request-flow work, API Catalog packaging, or operator access troubleshooting beyond clear handoffs.
kong-konnect MCP server first for live inspection of existing
meters, features, plans, subscriptions, and customer charges. It covers the
whole billing lifecycle, event ingestion and invoicing included, so search its
own tool index by name before concluding a step is impossible.kong-konnect MCP tools usually mean the client has not completed the
OAuth login for the server (or, for a headless PAT/SPAT configuration, the
bearer header is unset); treat it as a credential symptom and point to
docs/install/README.md.references/access-and-endpoints.md. It is
the one fact here that cannot be recovered by looking.terraform-konnect or kongctl-declarative path, confirm the
provider or CLI actually models these resources; otherwise keep changes on the
billing API and say so plainly.Load only the reference that matches the active branch:
references/access-and-endpoints.md
references/pricing-model-questions.md
references/catalog-entity-mechanics.md
references/usage-proof-and-invoicing.md
Establish which org you can write to before modeling anything. It is one call, and it decides whether this session ends in a published catalog or in a mapping the user has to build later.
Then list the meters, features and plans that are already there. Key collisions and near-duplicate features are the most common avoidable mistake, and an org that already carries a catalog usually needs a namespace prefix rather than the obvious names. Name the org back to the user before the first write.
Load references/access-and-endpoints.md for the credential bootstrap and the
endpoints.
Reduce the source — a page, a deal sheet, a competitor's tiers — to one table: per tier, the price and cadence, every included allowance with its unit, every overage rate, and every feature bullet.
When the source is a published pricing page, extract the tier table from the raw HTML yourself. Summarizers routinely mangle these pages: tabbed tier groups collapse into one another, tiers go missing, and numbers migrate between tabs. Pricing pages are usually tabbed with a full tier set per product line, and DOM order rarely matches tab order — identify each group by its own content, not its position.
A wrong number here reaches a published plan and bills a real customer. Confirm the tier count against what the source actually renders.
Before any entity exists, settle for each metered thing: which event marks it, what bounds it, and what makes it idempotent. "A conversation" or "a request" is not yet meterable.
Decide this jointly with the user when the source does not say. It determines the meter's aggregation and dimensions, and it is expensive to change once events are flowing.
Load references/pricing-model-questions.md and walk the model against it. For
each under-determined point, offer two or three concrete modelings with their
billing consequences, name your recommended default, and ask.
Batch the questions into one round so answering is confirmation rather than research. Record anything the user does not settle as an explicit stated assumption — never as a silent choice.
Before writing: the meter list, the feature list, one plan's full definition, a tier-to-rate-card table for the rest, the assumptions carried, and what you are deliberately not creating. Get approval.
Plans are versioned and publishing is a commitment. A wrong catalog is retired by archiving versions, not by editing them in place.
Meters, then features, then plans, then publish. Resolve ids by key at each step instead of hardcoding them, so a partial run can be repeated safely.
Load references/catalog-entity-mechanics.md for the entity contracts and the
price-type semantics that decide what actually gets charged.
A catalog that creates cleanly can still bill wrong. Subscribe a mock customer, seed a scenario whose arithmetic you can do by hand, and make the engine agree.
Choose volumes that cross an allowance boundary and include usage that is supposed to be excluded, so one number tests both the tiering and the filter.
Load references/usage-proof-and-invoicing.md for the seeding sequence and for
reading charges and invoices correctly.
State which plan, what usage, which tier it landed in, what the engine computed, and what will be billed when. Restate every assumption from step 4 that is still load-bearing.
value_property has to be. count ignores
it; sum, avg, min and max need one resolving to a number; unique_count
needs one resolving to a string. Decide it together with the aggregation — the
wrong pairing is rejected at create time.Before answering, verify that you can state:
konnect-ai-gateway when the real subject is AI Gateway request flow,
provider routing, or LLM analytics rather than billing the usage it produces.konnect-api-catalog or konnect-api-publish when the user is packaging
or publishing an API rather than pricing its consumption.konnect-access-scope when the blocker is token, role, org, or region
access rather than the billing model.konnect-platform-router when the request spans several Konnect surfaces
and the owner is unclear.