Competitor Profile
Where this is heading
Check is the counter-example that only shows up if the repository is read rather than the marketing. It ships the most permissive agent surface in this corpus and defaults it to a sandbox, with the only reversible payroll commit found anywhere in the set.
01
02
How the product hangs together, at the level a product leader needs to place it.
Check sells embedded payroll as an API. Its customers are software platforms that want to put payroll inside their own product rather than employers buying payroll directly, which puts Check underneath most tracked payroll vendors rather than beside them. It was incubated at Stripe and sells to developers, so the buying centre and the sales motion are both different from a platform sold to an HR or payroll function.
The agent surface is unusually complete. Check publishes a first-party MCP server exposing more than 260 operations, and the payroll set includes creating payrolls, adding earnings and deductions, previewing totals and approving runs. Mutation is the default posture rather than an extension of a read-only base, which makes it the most permissive surface found in this corpus by a distance.
03
Geographic footprint, and where payroll is native rather than partner-delivered.
North America: NATIVE for US payroll, delivered as infrastructure inside a partner's product rather than as a product a buyer chooses.
United Kingdom: NOT SERVED. No UK statutory payroll, so PAYE, RTI and CIS do not arise.
EU and South Africa: NOT SERVED in any source found.
The competitive question Check raises is not geographic. A platform that embeds Check gains payroll without building it, which changes who counts as a payroll competitor: the visible brand in a deal may be running someone else's engine, and a capability comparison that stops at the brand will misread the market.
04
Check sells to engineering and product teams rather than to payroll buyers, and its documentation and repository are the sales surface. No pricing is published in any published source, which is normal for infrastructure sold through partnership agreements and means a buyer-side price comparison is not available from outside.
The commercial terms that would matter most for the agent question are also the ones not published in marketing. Two documents would settle whether the production server is used against live payrolls by operations staff rather than by engineers: the terms of service, which would say who bears the loss on an agent-initiated approval, and the changelog, which would show the sandbox default changing.
05
The repository qualifies the marketing in four ways, and none of the four is visible in the product description.
First, the API base URL defaults to sandbox, so an agent picking the server up out of the box is not touching live payroll. Second, four of the fifteen payroll tools are simulation scaffolding that advance a payroll through its processing states with no real money movement. Third, the tool mode defaults to dynamic, so three meta-tools sit in the model's context rather than 263 individual ones, which means the advertised operation count is a catalogue size and not a context payload. Fourth, and most interesting, approve_payroll is paired with reopen_payroll.
That pairing is the finding. Every other agent-callable payroll commit found in this corpus is one-way. A commit that can be undone changes the risk calculation on autonomy completely, because the cost of a wrong action stops being a filed return and becomes a reopened one. The repository header reads Early Access Beta, available to select partners, so this is a design position rather than a deployed reality, and it is a design position worth watching.
This is the case for reading repositories rather than product pages. Every one of these qualifications lives in code and configuration defaults, and none of them appears in the marketing.
06
Diagnosis and the play together. Each is a live attack vector, not a general weakness.
Check does not sell to employers, so it cannot be beaten in a deal it never enters. Where it competes is one layer down, for the platform that would otherwise build payroll itself or embed a rival, and the argument against it there is jurisdictional: a platform with any ambition outside the United States gains nothing from an engine that files in one country.
The sandbox default and the Early Access Beta status are also a fair line of attack for any competitor whose agent surface is in production against live payrolls. A catalogue of 260 operations against a sandbox is a smaller claim than it reads as, and the honest comparison is production surface against production surface.
Public evidence on Check is materially thinner than on comparable vendors, and that thinness should be closed before this profile is cited as load-bearing evidence in either direction.
07
What Check has bought, where each sits now, and what it says about direction. None of these are signals; they are standing facts about who owns what.
No acquisitions found in the five-year window.
Checked 2026-08-15. Operating cadence and status are re-checked on each profile refresh; a change of state is itself a finding, because nobody announces the day an acquisition finishes being absorbed.
08
The latest signals where Check is the subject. Full history on the signals page.
09
The live predictions whose evidence rests on this vendor, including the ones going against us.