Three tools in a fixed order: list the templates, fetch the chosen template's field schema, then submit field data that matches it. The schema call is not optional — the fields are defined on the gotoHuman side, so an agent that guesses at them submits a review nobody can act on. What distinguishes this from a single approve-or-reject callback is the workflow argument: one runId plus prevSteps links several reviews, possibly by different people, into one traceable run. Templates themselves are built in gotoHuman, not here.
A server for human-in-the-loop review — list the review templates, read the fields one expects, and submit a review request against it.
- The review templates available on the account, and the exact fields and config a chosen template expects
- A review request carrying field data, a title, assigned users and metadata, so what a reviewer sees is structured rather than a wall of text
- Multi-step chains: the same runId on each step with prevSteps naming the reviews that came before, which ties a sequence of approvals into one run
- A webhook URL per request, for where the decision is delivered
GOTOHUMAN_API_KEY, and a gotoHuman account with at least one review template already defined — the field schema comes from the template, so it has to exist before an agent can ask for a review. MIT.
One command plus a key — npx @gotohuman/mcp-server, then supply credentials