report_checkout_step
Call at each step of your own checkout (checkout_by agent): the user watches it live in the Envoy app.
MCP tool report_checkout_step, called with tools/call on https://api.envoy.money/mcp (Streamable HTTP).
Call at each step of your own checkout (checkout_by agent): the user watches it live in the Envoy app. Moves no money, returns no card. Repeats are harmless.
| Title | Service | Read only | Destructive | Idempotent | External sites |
|---|---|---|---|---|---|
| Report a checkout step | purchase | false | false | true | false |
The text content is a plain-English briefing, then one line of JSON with the slim agent data. structuredContent holds that same slim data. isError is true when the call failed and changed nothing. Diagnostics for support are in _meta["money.envoy/support"]; they are not for the model. Tool annotations describe effects; they do not grant permission. The HTTP API returns the full envelope: see the API reference page.
Input
Prop
Type
Unknown fields are rejected. The exact input schema is on the schemas page.
Synthetic request
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "report_checkout_step",
"arguments": {
"purchase": "synthetic-intent-0001",
"step": "payment_page_reached"
}
}
}Success data
The text content opens with a briefing: one to three sentences for the user, then What to do: for the agent. Tell the user the first part in your own words and do what What to do says. One JSON line follows with these fields. Pass next.args unchanged to next.tool. Never show the JSON line, ids or codes to the user. Outputs may gain fields, so ignore ones you do not recognize.
| Field | Required | Type | Description and rules |
|---|---|---|---|
situation | No | "needs_details" | "needs_answer" | "waiting_for_user" | "working" | "ready" | "confirming" | "done" | "stopped" | "needs_support" | Where the purchase stands. The English briefing above says what it means. |
purchase_id | No | string | Pass as purchase in later calls. Never show it to the user. |
next | No | object | The call to make next: tool, args (pass verbatim) and wait_seconds before calling. |
retry | No | "same_call" | "after_fix" | "check_status_first" | "new_purchase" | "no" | Whether and how to retry. |
fix | No | array | Input fields to fix before retrying. |
support_ref | No | string | Give this to the user only if they ask for help. |
Synthetic result
Example synthetic-unauthenticated (synthetic data, the MCP text content; JSON is formatted here for reading but arrives on one line):
Envoy needs to be reconnected to this assistant.
What to do: Ask the user to reconnect Envoy, then try again. If the user asks for help, give them the reference EV-72Z0SF.
{
"retry": "no",
"support_ref": "EV-72Z0SF"
}Errors
A failure sets isError to true and changes nothing. The briefing says what happened and what to do; retry and fix in the JSON line say whether and how to retry. Branch on those, not on a code: these common codes appear only in _meta["money.envoy/support"], for support. See all error codes for the complete contract list.
| Code | HTTP status | Meaning | What the agent does |
|---|---|---|---|
INVALID_INPUT | 400 | Fix the supplied input. | Fix the fields listed in the data below and make the same call again. |
UNAUTHENTICATED | 401 | Reconnect through Central Auth. | Ask the user to reconnect Envoy, then try again. |
UNAVAILABLE | 503 | The service is unavailable; follow retryable and next. | Wait, then make the same call again. Don't pay any other way meanwhile. |