live wire
AI · Red Hat documents usage-based admission fair sharing for Kueue 1.4 on OpenShiftRed Hat DeveloperAI: Red Hat maps governed firewall changes from ServiceNow through Ansible and two human approval gatesRed Hat DeveloperCLUSTER MGMT · ACM 2.17 makes Submariner 0.24 GA with Important-rated fixesRed Hat ErrataPLATFORM · Red Hat makes on-premises Lightspeed recommendations GA for Satellite 6.18Red Hat ErrataSECURITY · Red Hat Hardened Images updates Tomcat 10 for nine authentication, access-control and DoS flawsRed Hat ErrataAI · Open Data Hub 3.6.0 EA1 bundles Trainer, MLflow and llm-d componentsOpen Data HubAI · Speculators 0.6.0 adds P-EAGLE parallel drafting for vLLM speculative decodingRed Hat DeveloperSECURITY · OpenShift 4.17.57 fixes seven Go and TLS CVEs in an Important-rated updateRed Hat ErrataAI · Red Hat benchmarks local LLM guardrails with EvalHub, exposing regex accuracy and latency trade-offsRed Hat DeveloperAI · Red Hat maps silent tool-call failures across agentic pipelinesRed HatAPI · Kuadrant 1.5.3 adds GRPCRoute policies and developer-portal API-key workflowsKuadrantAI · (Aug 25) IBM releases Apache-2.0 Granite 4.2 reasoning models in 3B, 8B and 30B sizesIBM ResearchJAVA · Red Hat build of Quarkus 3.33.3.SP1 fixes 13 CVEs in an Important-rated updateRed Hat errataAI · vLLM moves Kimi K2 RL weight sync across 384 H100s in 7.53 seconds (Aug 22)vLLMAI · Red Hat documents usage-based admission fair sharing for Kueue 1.4 on OpenShiftRed Hat DeveloperAI: Red Hat maps governed firewall changes from ServiceNow through Ansible and two human approval gatesRed Hat DeveloperCLUSTER MGMT · ACM 2.17 makes Submariner 0.24 GA with Important-rated fixesRed Hat ErrataPLATFORM · Red Hat makes on-premises Lightspeed recommendations GA for Satellite 6.18Red Hat ErrataSECURITY · Red Hat Hardened Images updates Tomcat 10 for nine authentication, access-control and DoS flawsRed Hat ErrataAI · Open Data Hub 3.6.0 EA1 bundles Trainer, MLflow and llm-d componentsOpen Data HubAI · Speculators 0.6.0 adds P-EAGLE parallel drafting for vLLM speculative decodingRed Hat DeveloperSECURITY · OpenShift 4.17.57 fixes seven Go and TLS CVEs in an Important-rated updateRed Hat ErrataAI · Red Hat benchmarks local LLM guardrails with EvalHub, exposing regex accuracy and latency trade-offsRed Hat DeveloperAI · Red Hat maps silent tool-call failures across agentic pipelinesRed HatAPI · Kuadrant 1.5.3 adds GRPCRoute policies and developer-portal API-key workflowsKuadrantAI · (Aug 25) IBM releases Apache-2.0 Granite 4.2 reasoning models in 3B, 8B and 30B sizesIBM ResearchJAVA · Red Hat build of Quarkus 3.33.3.SP1 fixes 13 CVEs in an Important-rated updateRed Hat errataAI · vLLM moves Kimi K2 RL weight sync across 384 H100s in 7.53 seconds (Aug 22)vLLM
upstreambeat.ai
newsAPI

Kuadrant puts A2A agent traffic inside its MCP Gateway policy perimeter

An experimental router path turns A2A agent and method metadata into trusted headers for existing authorization, rate-limit and telemetry policies.

Gateway policy path for A2A agent traffic
AI-generated diagram
By The News Desk· Sep 1, 2026the quick take — two AI hosts, this story only

Kuadrant has added an experimental Agent2Agent passthrough path to its MCP Gateway, giving platform teams a first way to apply the gateway’s existing policy controls to traffic between AI agents. The feature is off by default and deliberately stops short of agent discovery, task ownership or automatic routing.

What changed

The new --enable-a2a flag teaches the gateway router to inspect requests under an /a2a path and derive two headers: x-a2a-agent, taken from the first path segment after /a2a/, and x-a2a-method, taken from the JSON-RPC method in the request envelope. Kuadrant says the router strips client-supplied x-a2a-* headers before setting its own values, preventing callers from forging the metadata consumed by policy and telemetry.

Once those fields are ordinary trusted headers, existing Kuadrant resources can act on them. An AuthPolicy can authorize access per agent, a RateLimitPolicy can throttle agent traffic, and Telemetry configuration can include the addressed agent and A2A method in structured access logs. Unknown methods are normalized to other to bound metric cardinality, while malformed A2A requests are rejected rather than forwarded without policy metadata.

The implementation parses only the JSON-RPC envelope fields it needs, not the task payload. An operator still supplies the HTTPRoute and, where necessary, a URL rewrite from /a2a/{agent} to the endpoint exposed by the destination agent.

Who it affects

The immediate audience is platform teams running agents that delegate work to other agents while already using Kuadrant’s MCP Gateway for tool access. Kuadrant’s engineering post argues that tool calls may already pass through authentication, rate limiting and observability, while direct A2A delegation can otherwise bypass that policy perimeter.

This release slice does not provide a complete A2A control plane. It does not register or discover agents, serve agent cards, choose destinations, track task ownership or observe streamed task lifecycles. Kuadrant says those capabilities were proven in a fork but held back while their long-term API and placement remain unsettled.

What to do

Teams evaluating the feature should begin with the project’s A2A passthrough guide, enable the flag only in a test environment, and attach an HTTPRoute, AuthPolicy and Telemetry configuration. The practical validation is straightforward: confirm that callers cannot inject their own A2A headers, malformed envelopes fail closed, and access logs record the caller, destination agent, method and response outcome.

Because the feature is experimental and disabled by default, this is a lab path rather than a production migration signal. Its significance is narrower but concrete: Kuadrant can now bring inter-agent calls into the same authorization and audit machinery already used for MCP tool traffic.

Filed by The News Desk. Corrections: desk@upstreambeat.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.