@modelcontextprotocol/sdk 1.29.0. Muster now uses the spec-standard Mcp-Session-Id header, which @ai-sdk/mcp’s built-in HTTP transport handles natively.
See ./docs/releases/v0.132.1-changelog.md for more information.LatestOciReleaseProcessor that annotates Component entities carrying giantswarm.io/helmcharts with giantswarm.io/latest-release-tag and giantswarm.io/latest-release-date from the referenced OCI registry. For multi-chart entities the highest-semver stable tag wins; prerelease tags are skipped. Toggle via catalog.processors.latestOciRelease.enabled.@giantswarm/backstage-plugin-gs-node node-library package and move the container-registry client code into it so it can be shared between gs-backend and the catalog module. Move parseChartRef from plugins/gs to gs-common so it can be used backend-side.
See ./docs/releases/v0.132.0-changelog.md for more information.LatestReleaseProcessor and SbomDependencyProcessor work for repos outside the App’s installation (and public repos work without any integration configured).ai package to a single version via root yarn resolution to fix AI_TypeValidationError when invoking MCP tools.
See ./docs/releases/v0.131.1-changelog.md for more information.LatestReleaseProcessor that annotates Component entities with giantswarm.io/latest-release-tag and giantswarm.io/latest-release-date from GitHub Releases, with optional tag-prefix matching for monorepos.KlausProvider, a catalog entity provider that discovers Klaus personalities, toolchains, and plugins from GitHub and emits them as Component entities..Values.probes.{liveness,readiness} (initialDelaySeconds, periodSeconds, timeoutSeconds, failureThreshold) and the default timeoutSeconds is raised from the k8s implicit 1 to 5. The HTTP path/port stay fixed (/.backstage/health/v1/{liveness,readiness} on port). The k8s default 1s is tight against the chart’s 500m CPU limit: event-loop stalls during plugin startup, GC, or DB reconnects regularly miss the deadline and surface as Unhealthy: ... context deadline exceeded events even though the pod keeps serving traffic (observed on the BWI Backstage Deployment on spidertron, 2026-05-11).
See ./docs/releases/v0.130.0-changelog.md for more information.useChatSetup constructed a fresh AssistantChatTransport on every render and handed it to useChatRuntime; when its identity changed mid-stream (e.g. as a side effect of a state update triggered by streamed reasoning-delta / tool-input-delta events), the runtime tore down the in-flight chat fetch with TypeError: network error. Envoy logged the symptom as response_flags: DC (downstream remote disconnect) against a healthy Backstage upstream while the SSE stream summary was sawFinish=false, last event tool-input-delta, surfacing in the UI as a “Network error” banner even though no real network outage occurred. The transport is now memoised on the resolved API URL and the stable getHeaders / debugFetch callbacks, so a single transport lives for the component’s lifetime.AbortSignal and log abort events with the reason inline (ABORT signaled by client at <ms> -- reason: <name>: <message>). Stream-outcome lines now annotate aborted streams with [client-aborted at Nms reason="..."] so a “Network error” banner can be classified as a deliberate client cancel (transport rebuild, unmount, manual stop) versus a real proxy / network failure (no abort signal fired, raw stream error). Read errors that fired after the caller aborted are now logged as console.warn (“cancelled by client AbortSignal”) rather than console.error (“STREAM READ FAILED”), reserving the latter for the genuine pre-finish, non-aborted failure mode.See ./docs/releases/v0.129.3-changelog.md for more information.
bytes=... events=... sawFinish=... lastEventType=...) instead of attaching it as a trailing object argument, so it stays readable in browser-console consumers that flatten the args array (devtools-snapshotters, log shippers, the Cursor IDE browser, etc.) rather than collapsing to “[object Object]”. Also classify a stream read error that arrives after a finish event was already parsed as a post-completion teardown (console.warn, message already committed) instead of a network failure (console.error); this matches the actual semantics of the AI SDK aborting the underlying fetch once it has finished consuming the stream.
See ./docs/releases/v0.129.2-changelog.md for more information.ai-chat-verbose-debugging feature flag. Verbose payload-level logging (messages, system prompt, tool schemas, per-event SSE detail) remains gated on the feature flag in non-production builds.
See ./docs/releases/v0.129.1-changelog.md for more information.streamText() without temperature, topP, topK, seed, minP, or maxOutputTokens, so the server’s defaults applied – which for vLLM means temperature=1.0, top_p=1.0, top_k=-1, seed=null. That is far too loose for a tool-using agent backed by a reasoning model and was the dominant cause of token-cost variance in production agent loops (same prompt, fresh chat, observed total-token spread of 22k / 607k / 22k across three runs against the same Qwen3 endpoint). Config now accepts an aiChat.sampling block with temperature, topP, topK, minP, seed, and maxOutputTokens; all fields are optional and default behaviour with no sampling: block is unchanged. temperature, topP, topK, seed, and maxOutputTokens are forwarded through the AI SDK to every provider that supports them. minP is spliced into the request body via the OpenAI-compatible provider’s transformRequestBody hook, since vLLM accepts it as a top-level field but it is not part of the AI SDK call settings. The ai-chat-backend README documents recommended values per model family (Qwen3 thinking/non-thinking, Qwen3-Coder, GPT-4 / GPT-4o, Anthropic Claude).
See ./docs/releases/v0.129.0-changelog.md for more information.tools field in the request body is optional, so it arrives as undefined whenever the frontend does not register any client-side tools. The previous implementation called Object.entries(tools) unconditionally, which threw TypeError: Cannot convert undefined or null to object, returning a 500 from POST /api/ai-chat/chat on every request. The browser surfaced this as “network error” and the LLM never reached the tool-merge step, so MCP-provided tools (muster, prometheus, kubernetes, …), skill tools, getDate, and the context-usage tool were also unreachable from chat. frontendTools now accepts null/undefined and treats them as an empty registry.MCPClient per server for 30 minutes, but the cache was holding on to clients whose StreamableHTTP transport had already been torn down by muster (idle timeout, server reset, …). Tool calls then failed with MCPClientError: Attempted to send a request from a closed client until the TTL expired, surfacing in the browser as a “network error” banner. The cache now chains into the transport’s onclose callback and evicts the entry as soon as the connection drops, so the next request rebuilds the client cleanly. Tool execution also detects the same SDK error eagerly and marks the entry dead so a single failure – not a 30-minute window – triggers reconnection.
See ./docs/releases/v0.128.1-changelog.md for more information.