Request
Path parameters
The unique identifier of the published agent to invoke (e.g.
"stripe-agent").Body parameters
Structured input payload forwarded to the agent. Defaults to
null if omitted.Natural language instruction for the agent.
Optional A2A task identifier. Allows you to correlate this invocation with an existing task.
Optional A2A context identifier for multi-turn conversation state.
Identifier of the specific agent skill to invoke. Omit to let the agent select the appropriate skill.
Bearer token or API key passed directly to the agent. Use
auth_context_id instead to reference a stored credential.Reference to a stored auth context. The gateway decrypts and forwards the credential without exposing it in the request log.
ISO country code of the calling region (e.g.
"AU"). The gateway enforces this against the agent’s allowed_regions policy.Set to
true to acknowledge and proceed with agents flagged as requiring human approval (human_approval_required: true). Defaults to false.Maximum cost units you are willing to spend. The gateway rejects the call if the agent’s
cost_per_call_units exceeds this value.Optional payment settlement request forwarded to the agent.
Raw A2A envelope to forward verbatim to the agent. Use this to pass protocol-specific fields not covered by the other parameters.
Response
Returns anInvokeAgentResponse on success.
The agent that was invoked.
Execution status reported by the gateway (e.g.
"completed", "failed").Unique identifier for the execution receipt persisted by the gateway. Use this to retrieve the receipt later.
A2A task identifier returned by the agent. Omitted if the agent does not return one.
A2A context identifier. Omitted if not present.
Human-readable status message. Omitted if not present.
Structured output from the agent. Omitted if the agent returns no structured data.
Normalized settlement request echoed back from the gateway. Omitted if no settlement was requested.
Payment receipt from the settlement rail. Omitted if no payment occurred.
The raw A2A JSON-RPC response returned by the agent endpoint.
Status codes
| Code | Meaning |
|---|---|
200 OK | Invocation completed. Check status in the response body for the execution outcome. |
403 Forbidden | Policy check failed: agent is blocked, provider is revoked/blocked, region not allowed, cost limit exceeded, or confirm_risky not set for a high-risk agent. |
404 Not Found | No published agent with the given agent_id exists. |
502 Bad Gateway | The gateway could not reach or received an error from the agent’s downstream endpoint. |