Supernova Review (2026): Pricing, Features & Honest Verdict
Supernova is a data platform that syncs your business tools into an encrypted Iceberg lake and exposes the whole thing to Claude or Codex over MCP. Best for seed and Series A teams with no data engineer. Price: usage-based from $0.15 per GB-hour of compute, no free plan. Rating: 7.0/10.
What is Supernova?
Supernova connects your live business data to an AI assistant without a warehouse, a BI tool or a data team standing in between. You point it at Stripe, HubSpot, Postgres or any of about 30 other sources, it syncs them into an encrypted Apache Iceberg lake in object storage, and then you add mcp.supernova.ai/mcp to Claude or Codex and start asking questions. The pitch on the homepage is a churn analysis answered in a chat window: eleven accounts, $63,400 of ARR, nine of them lost to a failed payment retry.
The company has been at this since 2020 and launched on Product Hunt on 21 August 2026, finishing second for the day. The legal entity is Infinite Canvas Inc., a Delaware corporation trading as Supernova AI from an address on Market Street in San Francisco, which we found in the terms of service rather than the footer. Worth a quick disambiguation: this is supernova.ai, and it has nothing to do with supernova.io, the design system platform that shares the name.
What are Supernova’s key features?
An MCP endpoint that is the whole product
One URL added to Claude Desktop or Codex makes every synced table queryable in about a minute. This is the feature everything else exists to serve, and it is the reason the product is interesting now rather than in 2022. Seats are free and unlimited, and MCP access is included at no charge, so the marginal cost of giving a fourth person access to company data is zero. Most BI vendors charge per viewer, which is exactly the pricing that keeps data locked in one analyst’s head.
The lake, the catalog and the engine in one
Supernova bundles the pipeline, the storage and the query engine, so there is no Snowflake or BigQuery bill underneath it. Tables live as open Iceberg in object storage with table-level encryption, every table keeps version history for time travel, and zero-copy connections mean anything that speaks Iceberg can read the lake directly. That last point is the anti-lock-in argument and it holds up: if you leave, the data is already in an open format you can point another engine at.
TypeSQL and the CLI
Supernova ships a typed SQL layer that autocompletes across joins and type-checks before a query runs, so a misspelled column is caught with a suggestion instead of an error at runtime. Columns carry semantic annotations too, so a currency amount stored as 4812000 renders as $48,120.00 and a country code renders as a flag. The CLI is a single binary that connects sources, runs queries, tails live changes and wires up Claude, and it works from a laptop or from CI.
Connectors, including ones that do not exist yet
Postgres and MongoDB sync in real time. Stripe, HubSpot, Salesforce, Shopify, NetSuite, MySQL, the ad platforms, Intercom, Zendesk, Jira, Mixpanel, Amplitude and about twenty more are on the list. Schema history is tracked and syncs are monitored from a command centre view. If your source is not covered, the site commits to building a custom connector in 24 hours, which is a specific promise rather than a vague one.
How much does Supernova cost?
Supernova is pure usage pricing with no plans. Compute is $0.15 per GB-hour for query and sync, billed by the second. Storage is $0.05 per GB-month, egress is $0.10 per GB, read operations are $0.40 per million, and write and list operations are $5 per million each. Seats cost nothing and there is no user limit. AI tokens are charged at 1.5 times pass-through of the model’s list price.
There is no free tier and no trial published anywhere on the site, the docs or the pricing page. The FAQ says everyone is on the premium plan and there are no premium plans, which is a nice line, but it also means the cheapest way in is a credit card. Compare that to Airbyte, which has a $0 plan and an always-free self-managed edition, or MotherDuck, which starts at $0 with 10 GB of free storage.
Why does Supernova’s pricing page contradict itself?
Here is the problem, and it is on the pricing page’s own first line. Supernova says: “Two meters, rows moved and compute, and everything else is included.”
The rate card directly underneath that sentence lists nine line items. Compute, storage, egress, read operations, write operations, list operations, seats, Claude and Codex access, and AI tokens. Six of those nine carry a real charge. So “everything else is included” is contradicted by the table it introduces.
Worse, rows moved is named as one of the two meters and no rate for it appears anywhere on the page. Then the homepage FAQ answers “How does the pricing work?” with “There are no per-user or per-row charges.” So the pricing page says rows are a meter, the homepage says rows are not charged, and neither publishes a row rate. Three statements, no consistent answer, about the meter the vendor puts first.
This matters more than it might sound, because rows are how the category prices itself. Fivetran bills on Monthly Active Rows, publishes a calculator and a worked example totalling $549.36 a month, and gives 500,000 MAR free. A buyer arriving from Fivetran will read “rows moved” and reasonably expect a comparable number. There is not one. The homepage summary also drops egress and list operations entirely and shows only five meters, so the two pages do not even agree on how many charges exist.
What does the 1.5x AI token markup cost you?
Supernova charges 1.5 times the model’s list price for tokens consumed inside the product. Credit where it is due: it says so, on the rate card, as a number. Most vendors that resell inference describe a “small markup” and never quantify it, and we have written up several that never state one at all.
But quantified is not the same as cheap. A 50 percent markup on inference is at the high end of what anyone charges. OpenRouter, the usual public benchmark, takes no markup on inference at all and makes its money on a 5.5 percent credit-purchase fee. If your workload leans on the built-in AI modelling rather than on your own Claude subscription through MCP, that multiplier is the line item to watch.
The good news is that you can route around it. Because MCP access is free and seats are free, querying through your existing Claude or Codex subscription costs you nothing extra on the AI side, and you pay Supernova only for compute and storage. Use the platform as the lake and let your own assistant do the thinking.
| Plan | Price | Plan Features | Best For |
|---|---|---|---|
| Compute | $0.15/GB-hour | Query and sync, billed by the second | The main meter on any workload |
| Storage | $0.05/GB-month | Encrypted Iceberg lake in object storage | Keeping history and time travel |
| Operations | $0.40 per million reads | Writes and list operations are $5 per million each; egress $0.10/GB | High-frequency query patterns |
| Seats and MCP | $0 | Unlimited users, no per-seat fee, MCP access included | Giving the whole team access to company data |
| AI tokens | 1.5x list price | Pass-through of the model's list price, metered as used | Built-in modelling; route via your own Claude plan to avoid it |
Who is Supernova best for?
Use Supernova if you are a startup between seed and Series B with revenue data in Stripe, customer data in HubSpot and product data in Postgres, no data engineer, and a founder who wants answers in Claude instead of a dashboard request queue. The free unlimited seats mean the whole team can ask, and the usage pricing means a quiet month costs very little.
Use it if you want to avoid the Fivetran plus warehouse plus BI tool stack, which is three vendors and three bills.
Skip Supernova if you need to model your bill before you commit, because the headline meter has no published rate. Skip it if you require a free tier to evaluate, because there is not one. And skip it if your data lives mostly in sources outside the connector list, unless you are willing to test the 24-hour custom connector promise.
Best Supernova alternatives
Fivetran is the incumbent for the sync half. It has a free plan covering 500,000 monthly active rows and 3,500 activation rows, then charges by MAR with a public estimator. You still need a warehouse and a BI tool underneath, so the total stack costs more, but you can price it exactly before you buy.
MotherDuck is the closest thing on the query side, with its own MCP support. Lite starts at $0 with 10 GB of free storage, Business is $250 per org per month plus usage, storage is $0.04 per GB-month and compute starts at $0.60 an hour billed per second. Note the units differ from Supernova’s GB-hour compute meter, so the two are not directly comparable without modelling your own workload.
Airbyte is the open-source route. The self-managed edition is always free, the cloud Free plan is $0, Individual is $29 a month with overage at $0.004, and Team is $299 a month with overage at $0.005. More setup work, but the only option here you can run entirely on your own infrastructure at no licence cost.
Final verdict: is Supernova worth it?
The product is good and the ambition is real. Bundling the pipeline, the lake and the engine, keeping the data in open Iceberg so you can walk away with it, charging nothing for seats or MCP access, and answering a churn question in a chat window instead of a ticket, is a coherent answer to a stack most small teams cannot staff. The investor list, which runs from Sequoia through operators at Snowflake, Looker, dbt and Monte Carlo, suggests people who know this category think so too.
The pricing copy is what holds the score at 7.0. A page headed “Transparent pricing” that names a meter it does not price, claims two meters while listing six chargeable ones, and is contradicted by its own homepage FAQ, is not transparent. The rates that are published are clear and reasonable, so this reads as sloppy writing rather than anything worse. Ask a founder what a row costs before you sync anything large, and if they answer plainly, this is an easy tool to recommend.
Supernova Pros & Cons
What We Like
- Seats and MCP access both cost nothing, so the marginal cost of giving another person access to company data is zero
- Data sits in open Apache Iceberg with table-level encryption, so zero-copy connections let any Iceberg engine read it and you can leave without an export
- Bundles the pipeline, the lake and the query engine, so there is no separate Snowflake or BigQuery bill underneath
- The 1.5x AI token markup is at least stated as a number, which is more than most vendors reselling inference manage
- Around 30 connectors with real-time Postgres and MongoDB sync, plus a commitment to build a custom connector in 24 hours
What Could Be Better
- The pricing page opens with 'two meters, rows moved and compute' and then lists nine line items, six of which carry a charge
- No rate for 'rows moved' is published anywhere, while the homepage FAQ separately says there are no per-row charges. Three statements, no consistent answer
- No free tier and no trial published on the site, the docs or the pricing page. Airbyte and MotherDuck both start at $0
- The 1.5x token markup is a 50 percent premium, against OpenRouter which takes no markup on inference at all
- The About page lists more than 40 investors and zero employees. No founder is named anywhere on the site despite two calls to 'talk to a founder'
Supernova FAQ
What is Supernova?
Supernova is a data platform that syncs sources like Stripe, HubSpot and Postgres into an encrypted Apache Iceberg lake, then exposes them to Claude or Codex through an MCP endpoint. It bundles the pipeline, the storage and the query engine, so you do not need a separate warehouse or BI tool. Note this is supernova.ai, not supernova.io the design system platform.
How much does Supernova cost?
Pricing is pure usage with no plans. Compute is $0.15 per GB-hour billed by the second, storage is $0.05 per GB-month, egress is $0.10 per GB, read operations are $0.40 per million, and writes and list operations are $5 per million each. Seats and MCP access are free. AI tokens are charged at 1.5 times the model's list price.
Is Supernova worth it?
For a startup with no data engineer, yes. Free unlimited seats, open Iceberg storage with no lock-in and a working MCP endpoint make it a real alternative to a three-vendor stack. The catch is that you cannot fully model your bill in advance, because the meter the pricing page names first has no published rate. Ask what a row costs before syncing anything large.
What are the best Supernova alternatives?
Fivetran covers the sync half with a free plan up to 500,000 monthly active rows and a public MAR estimator, though you still need a warehouse. MotherDuck starts at $0 with 10 GB free storage, $250 per org per month for Business, and supports MCP. Airbyte is the open-source route, always free self-managed, $29 a month for Individual and $299 for Team.
Does Supernova offer a free plan?
No. We could not find a free tier or a trial on the pricing page, the homepage or the docs. The FAQ says everyone is on the premium plan and there are no premium plans, which means usage billing from the first query. Seats and MCP access are free of charge, but access to the platform itself is not.
Who is Supernova best for?
Startups between seed and Series B with revenue in Stripe, customers in HubSpot and product data in Postgres, no data engineer, and a founder who would rather ask Claude than file a dashboard request. Because seats cost nothing, it suits teams where several non-technical people need to ask questions of the same data.






