Target

Field reference for the Target configuration: the upstream destination configuration of an Agent Surface.

The target section defines the single upstream destination the gateway forwards traffic to: the actual managed agent backend. Every surface has exactly one target.

Fields

The target is shown in the canvas as the Managed Agent node.

FieldTypeRequiredDefaultDescription
Target Endpoint URLstringYes—Upstream destination URL. Accepts https://, fabric://{gateway_id}/{surface_id} for gateway-to-gateway routing, proxy://{id} for MCP proxy targets, or a2a-proxy://{id} for A2A proxy targets.
Endpoint authTargetAuthConfigNoNoneService-level credentials injected by the gateway into requests forwarded to the target (gateway → target authentication).
Policy Definition (request)Policy definitionNoNoneRequest-side OPA policy. Evaluated after the access point pipeline. Answers: “is this caller allowed this operation?”
Policy Definition (response)Policy definitionNoNoneResponse-side OPA policy. Evaluated on the response from the target before returning to the caller. Answers: “is this response allowed for this caller?”
PaymentPaymentPolicyNoNonePayment requirements. Tagged union: x402 (HTTP 402 protocol) or mpp (Machine Payments Protocol).
Per-tool policy bindingsMcpToolPolicyEntry[]No[]Per-tool RBAC entries for a Managed Agent targeting an MCP Proxy, set on the node labelled MCP Proxy. A tool with no matching entry falls back to the panel’s Default Tool Policy if one is set, and is otherwise denied whenever the entry list is non-empty. See Per-tool policy bindings.
NetworkingNetworkingNoNoneTimeout, retry, circuit breaker, and traffic mirroring configuration.
Agent IdentityIdentity injectionNo{}Controls whether and how a signed Verifiable Presentation is injected into outbound requests.
Trust RegistryTrustRegistryInjectionNoNoneInjects a trust registry extension into outbound requests forwarded to the target.
Trust Registry (inbound)TrustRegistryInjectionNoNoneValidates trust registry extension on inbound responses from the target.
Extension ValidationExtensionRulesNoNoneResponse extension schema validation (response from target).
Custom MetadataCustomMetadataNoNoneCustom metadata injected into request bodies forwarded to the target.
Custom Metadata (response)CustomMetadataNoNoneCustom metadata injected into response bodies returned to the caller.
Trust CheckTrust Check elementsNo[]Per-element TRQP verification on the target leg (Managed Agent → Target). Results appear as input.trust_check_results.target in OPA evaluation.
via MCP ProxystringNoNoneMCP proxy backend ID. Required when endpoint uses the proxy://{id} scheme.
via A2A ProxystringNoNoneA2A proxy backend ID. Required when endpoint uses the a2a-proxy://{id} scheme. Shown for A2A surfaces only.
Display namestringNoNoneDisplay name shown in the dashboard for fabric:// endpoints. Used when the remote gateway is offline.
MPP Auto-PayboolNofalseAutomatically pays MPP challenges for fabric:// targets without caller involvement.
Max amount per requeststringNoNoneSafety cap on the amount per auto-pay request. Expressed as a decimal string (for example "0.01").

Endpoint schemes

SchemeFormatUse case
HTTPShttps://<host>:<port>/<path>Standard HTTP/S upstream target.
Fabricfabric://{gateway_id}/{surface_id}Gateway-to-gateway routing through the fabric.
Proxyproxy://{mcp_proxy_id}Routes through a configured MCP proxy backend. Requires mcp_proxy_id.
A2A Proxya2a-proxy://{a2a_proxy_id}Routes through a configured A2A proxy backend. Requires a2a_proxy_id. Available for A2A surfaces only.

Networking

Controls resilience behaviour for requests forwarded to this target.

FieldTypeRequiredDescription
TimeoutsTimeoutConfigNoConnect and read timeout settings in milliseconds.
RetryRetryConfigNoRetry policy: max attempts, backoff strategy, and retryable status codes.
Circuit BreakerCircuitBreakerConfigNoCircuit breaker thresholds: failure rate, open duration, and half-open probe count.
Traffic MirroringMirrorConfigNoTraffic mirroring: duplicate requests to a secondary endpoint for testing or analysis.

MCP tool policies

Per-tool OPA policy binding. Evaluated whenever mcp_tool_policies is non-empty, which enables allowlist enforcement automatically.

FieldTypeRequiredDefaultDescription
Tool namestringYes—Name of the MCP tool this policy applies to. Must match the tool name exactly.
Policy DefinitionstringYes—ID of the OPA policy definition in the policy store.
DescriptionstringNo""Human-readable note shown in the dashboard.

Identity injection

Controls signed Verifiable Presentation injection into outbound requests.

FieldTypeRequiredDefaultDescription
Send signed identity VP to targetboolNotrueWhen enabled, the gateway signs and injects a VP into each outbound request to the target. Disable it if the target should receive the request without an injected VP.
Identity extraction strategystringNoNoneHow the agent DID is extracted. Options: From Payload, From API Key, From mTLS certificate, Static DID, From JWT Claims.
API KeystringNoNoneAPI key identifier. Required when strategy is From API Key.
CertificatestringNoNoneCertificate identifier. Required when strategy is From mTLS certificate.
DIDstringNoNonePre-known DID. Required when strategy is Static DID.
JWT claimstringNooidJWT claim to derive the DID from. Used when strategy is From JWT Claims.
Namespace claimsstring[]No[]Additional claims used to namespace the derived DID (for example iss, tid). Used when strategy is from_jwt_claim.
Payload fieldstringNoNoneTop-level payload field the identity payload is read from.
Dot-notation pathsstring[]No[]Dot-notation paths concatenated and hashed to derive the agent DID. Used when strategy is from_payload.
JSON SchemamapNoNoneJSON Schema describing the expected identity payload on the wire.