Skip to content

Buyer Guides

Red Hat's MCP Gateway GA: Choose Protocol and Feature Scope Together

Red Hat's October 16 gateway GA announcement leaves protocol, session features and console maturity distinct. Assess the combination your client flow needs.

Leviathan SystemsPublished 2026-10-205 min read

Red Hat announced the Model Context Protocol (MCP) gateway as generally available in Connectivity Link 1.5 on October 16, 2026. The same announcement labels progressive tool discovery stateful-only and the console plug-in developer preview. [1] For a buyer, those distinctions make the required client flow more informative than a product-wide maturity label.

Choose three things together: the product release, the protocol and session behavior for the intended flow, and the maturity of each required feature. Evidence for each item separately can still leave their combination unsupported. This matters when an application depends on a particular discovery behavior rather than simply reaching a gateway.

A selected working set needs session state

Red Hat describes discover_tools returning metadata without schemas; select_tools then narrows a stateful session's tool set and signals a list change. [1] Discovery tells an agent what it might choose. Selection determines the detailed tool definitions presented for its current work.

The dependency on state follows from that sequence. An earlier choice must remain associated with the session so a later listing can reflect it. Merely finding a tool by name does not supply that relationship. A buyer asking for this discovery experience therefore needs to identify where the selection is retained and which client interaction uses it. That is an explanation of the announced design, not a rule that every MCP discovery implementation must use the same arrangement.

Consider an original hypothetical example: an agent has 60 tools eligible under an unchanged authorization policy and selects six schemas for its current session. The selected definitions are a working subset of that eligible catalog. The example stipulates neither a permission change nor execution of any tool. Omitting the other 54 schemas from the current view does not establish that their underlying permissions were revoked.

Six out of 60 is one tenth of the tool count, but it is not a measured fraction of prompt tokens. Schema sizes, other context and the client's handling are unspecified. The useful conclusion is narrower: the current working set and the authorized set describe different things.

Protocol support does not extend every feature

The announcement names stateless protocol 2026-07-28 alongside 2025-11-25, 2025-06-18 and 2025-03-26, with negotiation on one endpoint. [1] That is a bounded vendor compatibility statement. It does not establish every client/server/feature combination.

For an evaluation, identify the actual client, upstream servers, negotiated protocol and required behavior along that flow. Reaching an endpoint or reporting a transport stream count leaves the additional discovery question unanswered. Ask the product provider to connect the proposed combination to support evidence for the offered release.

One client flow, two explicit requirement changes

Consider a fictional enterprise team evaluating Connectivity Link 1.5. Its original request has three mandatory requirements in one client flow: stateless communication, progressive tool selection retained as a session working set, and a console plug-in with GA status. This is a comparison of requirements with the announcement, with no installed environment or performed product test.

The announced scope does not meet that request. Progressive selection is stateful-only, so the stateless requirement conflicts with the requested discovery behavior. Separately, the console is developer preview, so it does not satisfy the GA-console requirement. The purchasing conclusion is to leave this bundle unaccepted: the gateway's GA headline resolves neither conflict.

In the first revision, the team replaces the stateless requirement with a stateful client flow and retains both progressive selection and the GA-console requirement. That removes the stated discovery conflict because the revised flow now uses the announced session-based path. It leaves the console conflict intact. Even a successful demonstration of stateful selection would not make the preview console GA; an offer relying on that console still falls short of the revised request.

In the second revision, the team keeps stateful selection but explicitly permits the announced developer-preview console. Only the console maturity requirement changes at this step. Neither of the two announced limitations now contradicts the stated requirements, so the combination becomes a candidate for detailed evaluation. That is a different purchasing position from accepting the original bundle, and a different assurance choice from requiring a GA console.

The next decision is whether evidence for the offered 1.5 release establishes the revised flow's suitability with the team's actual clients and upstream servers. The announcement alone cannot establish runtime compatibility, performance or security for that environment. Compare suppliers against the revised requirements; a demonstration of this stateful flow would still leave the original stateless request unanswered.

Protocol compatibility, tenant isolation, authorization and content screening answer different questions. The announcement's NeMo content checks are independent of authentication and authorization. [1] Selecting a smaller tool view therefore cannot establish either an access decision or all the controls needed around a request.

Use the JWT audience, time and permission guide when examining that access decision. Schema visibility alone does not establish permission. Conversely, permission to use a tool does not establish that the client can obtain its schema through the discovery flow it requires.

A tool-call audit event can precede execution

Red Hat says audit entries for valid tools/call requests cover calls reaching a backend and early router rejections. [1] An event can establish that the gateway observed and classified a request without establishing that the backend tool ran.

For a business-outcome claim, connect the gateway request to the downstream action and its result. A recorded attempt, backend receipt and intended business change are successive questions. If an offer also reports a success percentage, define which requests enter the count; the trace sampling and request metrics guide explains why retained records and all-request populations can differ. A count of audit events alone answers neither the execution nor the outcome question.

Content screening likewise leaves the access decision and the correctness of a permitted business operation to their own evidence. A request can satisfy a content policy while the intended business change remains unestablished.

Match the evidence to the offered release

The October 16 announcement establishes what Red Hat announced for Connectivity Link 1.5. Release-specific operating documentation and support evidence serve a different purpose: identifying how the offered combination is supported and what limitations apply. Documentation labeled 1.4 cannot establish 1.5 runtime requirements. [2] For the revised case, the remaining question is evidence for the chosen stateful flow and accepted preview feature, rather than another product-wide maturity label.

Connect a separate equipment need to physical deployment

If the selected infrastructure also needs equipment receiving, rack integration, cabling or physical handoff, Leviathan's Complete Physical Deployment service is the relevant scope. [3] Bring a nonconfidential equipment, site and handoff outline to the physical project inquiry. Gateway software evaluation and operation remain with the appointed software teams.

Contact

Discuss Your Infrastructure Project

Tell us about your project, and our team will follow up to discuss the scope, requirements and availability.