API quality is uncorrelated with vendor size
Some of the best APIs we work with belong to small companies and some of the worst belong to large ones. Xero and Stripe are excellent; several enterprise products are considerably harder.
Reference · updated August 2026
Integration marketplaces list logos. What a buyer actually needs to know is which APIs are stable, which paginate badly, which break on version changes, and which require a Windows machine somebody has to maintain. This is that list.
Tell us what you run. We will confirm what connects and what a custom integration would cost.
Systems with production connectors. Anything not listed is scoped case by case after a short paid discovery.
What we found
Patterns that hold across sixty-odd integrations, several of which contradict the assumptions buyers usually bring.
Some of the best APIs we work with belong to small companies and some of the worst belong to large ones. Xero and Stripe are excellent; several enterprise products are considerably harder.
A full historical extract from several accounting APIs takes days rather than hours, entirely because of rate limiting. It is the most common reason an integration schedule slips.
Duplicate transactions from a redelivered webhook are the most common problem we find in inherited integrations, and they are tedious to unwind because each duplicate is individually valid.
QuickBooks Desktop has no cloud API and is entirely workable through the SDK. It is roughly three times the effort, which is why we quote it separately rather than pretending otherwise.
Bidirectional writes require a designated authoritative side for every field. Where neither system is clearly authoritative, we recommend against it rather than building a conflict resolver.
Building a connector is one-off. Keeping it working through API versions, schema changes, and credential rotations runs $150 to $500 a month, and it is the line most proposals omit.
Any competent engineer can call an API and store the results. What makes an integration trustworthy is a daily reconciliation — counts, control totals, and spot-checked records compared between source and destination — that is written independently of the sync code.
Independence matters. A reconciliation that shares assumptions with the process it checks will confirm those assumptions rather than test them, and the failure it was built to catch is exactly the one it will miss.
QuickBooks Desktop, for the reasons above. Older on-premise ERP where the only path is a database read replica and the schema is undocumented. Industry-specific applications whose vendors treat customer data as a commercial asset — that last category is the only one we sometimes cannot solve, and it is a commercial obstacle rather than a technical one.
Marketplace settlement APIs deserve a specific mention. Amazon in particular returns settlement in a structure that requires substantial work to decompose into charges, fees, refunds, and reserves, and most integrations we inherit have simply not done it.
The APIs we would hold up as examples share four properties: stable identifiers that survive edits, a reliable way to ask what changed since a timestamp, bulk access that does not require pagination through everything, and version deprecation announced with real notice.
Xero, Stripe, Gusto, and Shopify meet most of that. The systems that are hardest to work with usually fail on the second property — no change feed — which forces a full comparison on every sync and makes the rate limits bind much harder.
An integration that cannot justify its maintenance. If a system is used monthly by two people and connecting it would cost $18,000 plus $400 a month, a scheduled export is the better answer and we will say so.
We turn this work down regularly. A neglected integration silently produces wrong numbers, which is worse than the manual process it replaced, and neglect is what happens to integrations nobody can justify maintaining.
This catalogue lists systems with connectors running in production for at least one customer. Nothing appears here on the basis of being technically possible or on a roadmap.
Maintenance cost figures are our actual internal cost of keeping connectors working, averaged over 2025 and 2026, rather than a price list. What we charge is stated on the services pages.
API quality assessments are our engineering team’s judgement from working with them, not a scored methodology. They are opinions held with some confidence and we would revise them if a vendor improved.
Anything not listed is scoped case by case through a paid discovery of $1,500 to $3,000, which produces a written feasibility report and a fixed quote whether or not you proceed.
Questions
We will confirm what connects, what needs building, and what is not worth connecting at all.