chore(formatting): undo shared formatter and keep structuredcontent

packages/mcp-core/src/api-client/client.ts

@@ -4282,8 +4282,7 @@ export class SentryApiService {
     opts?: RequestOptions,
   ): Promise<Event> {
     const body = await this.requestJSON(
-      apiPath`/organizations/${organizationSlug}/issues/${issueId}/events/${eventId}/` +
-        `?llmFormat=json`,
+      apiPath`/organizations/${organizationSlug}/issues/${issueId}/events/${eventId}/`,
       undefined,
       opts,
     );
@@ -5248,8 +5247,7 @@ export class SentryApiService {
     opts?: RequestOptions,
   ): Promise<AutofixRunState> {
     const body = await this.requestJSON(
-      apiPath`/organizations/${organizationSlug}/issues/${issueId}/autofix/` +
-        `?llmFormat=markdown`,
+      apiPath`/organizations/${organizationSlug}/issues/${issueId}/autofix/`,
       undefined,
       opts,
     );

packages/mcp-core/src/api-client/schema.ts

@@ -1178,8 +1178,6 @@ const BaseEventSchema = z.object({
   _meta: z.unknown().optional(),
   // dateReceived is when the server received the event (may not be present in all contexts)
   dateReceived: z.string().datetime().nullish(),
-  // shared-formatter output, present when the event endpoint is called with ?llmFormat
-  formatted: z.object({ format: z.string(), content: z.string() }).optional(),
 });
 
 export const ErrorEventSchema = BaseEventSchema.omit({
@@ -1414,8 +1412,6 @@ export const AutofixRunStateSchema = z.object({
     })
     .passthrough()
     .nullable(),
-  // shared-formatter output, present when the autofix endpoint is called with ?llmFormat
-  formatted: z.object({ format: z.string(), content: z.string() }).optional(),
 });
 
 /**

packages/mcp-core/src/internal/formatting.ts

@@ -43,7 +43,6 @@ import {
   getAutofixArtifactSummaries,
   getStatusDisplayName,
   isTerminalStatus,
-  wrapSeerContent,
 } from "./tool-helpers/seer";
 import { formatToolCallInstruction } from "./tool-helpers/tool-call-formatting";
 import { isPlainObject } from "./type-guards";
@@ -181,15 +180,6 @@ export function formatFrameHeader(
   }
 }
 
-/**
- * Whether the shared formatter covers this event type, and so whether its body should be used
- * instead of the local rendering.
- *
- * "default" is an error event without exception data, "generic" a performance regression or
- * metric issue, "csp" a Content Security Policy violation. Anything else (a transaction, most
- * notably) keeps the local path, which renders things the shared body does not carry such as
- * the fetched performance trace.
- */
 /**
  * Whether to read the issue's metadata instead of its top level fields. Performance issues can
  * have various categories such as 'db_query', but the issueType starts with 'performance_'.
@@ -208,10 +198,12 @@ export function isPerformanceIssueType(issue: {
   );
 }
 
-export function usesSharedFormatterBody(event: { type?: unknown }): boolean {
+/** Whether this server can render the event type; anything else arrives as an UnknownEvent. */
+export function isSupportedEventType(event: { type?: unknown }): boolean {
   return (
     event.type === "error" ||
     event.type === "default" ||
+    event.type === "transaction" ||
     event.type === "generic" ||
     event.type === "csp"
   );
@@ -1954,24 +1946,17 @@ function formatSeerSummary(autofixState: AutofixRunState | undefined): string {
     parts.push("");
   }
 
-  // Prefer the shared formatter's analysis for the body when the endpoint provides it, but
-  // keep the status handling around it: a run that failed or needs input must say so either
-  // way. Seer content is LLM-generated, so wrap it in the untrusted-data boundary.
-  if (autofixState.formatted?.content) {
-    parts.push(wrapSeerContent(autofixState.formatted.content, autofix.run_id));
-  } else {
-    // Summarize from the run's artifacts: the solution if available, otherwise
-    // the root cause if it has been identified.
-    const { rootCause, solution } = getAutofixArtifactSummaries(autofix);
-    if (solution) {
-      parts.push("**Summary:**");
-      parts.push(solution);
-    } else if (rootCause) {
-      parts.push("**Root Cause Identified:**");
-      parts.push(rootCause);
-    } else if (!isTerminalStatus(autofix.status)) {
-      parts.push("Analysis has started but no results yet.");
-    }
+  // Summarize from the run's artifacts: the solution if available, otherwise
+  // the root cause if it has been identified.
+  const { rootCause, solution } = getAutofixArtifactSummaries(autofix);
+  if (solution) {
+    parts.push("**Summary:**");
+    parts.push(solution);
+  } else if (rootCause) {
+    parts.push("**Root Cause Identified:**");
+    parts.push(rootCause);
+  } else if (!isTerminalStatus(autofix.status)) {
+    parts.push("Analysis has started but no results yet.");
   }
 
   if (autofix.status === "error") {
@@ -2132,14 +2117,7 @@ export function formatIssueOutput({
   // Event type union is: ErrorEvent | DefaultEvent | TransactionEvent | GenericEvent | CspEvent
   // But in practice we may have other types returned as UnknownEvent
   const eventType = event.type;
-  const isUnsupportedType =
-    eventType !== "error" &&
-    eventType !== "default" &&
-    eventType !== "transaction" &&
-    eventType !== "generic" &&
-    eventType !== "csp";
-
-  if (isUnsupportedType) {
+  if (!isSupportedEventType(event)) {
     // Log to Sentry for tracking new/unknown event types
     const sentryEventId = logIssue(
       `Unsupported event type encountered: ${String(eventType)}`,
@@ -2170,13 +2148,15 @@ export function formatIssueOutput({
 
     if (aiConversations && aiConversations.length > 0) {
       output += "\n## Response Notes\n\n";
-      output += formatAIConversationResponseNote({
+      output += buildAIConversationResponseNotes({
         aiConversations,
         organizationSlug,
         experimentalMode: experimentalMode ?? false,
         availableToolNames,
         directToolNames,
-      });
+      })
+        .map((note) => `- ${note}\n`)
+        .join("");
     }
 
     // For unsupported event types, return early without trying to render event details
@@ -2185,8 +2165,13 @@ export function formatIssueOutput({
 
   output += `**Event ID**: ${event.id}\n`;
   output += `**Type**: ${event.type}\n`;
-  const isSharedFormatterType = usesSharedFormatterBody(event);
-  if (isSharedFormatterType) {
+  // default: error without exception data, generic: performance/metric issue, csp: CSP violation
+  if (
+    event.type === "error" ||
+    event.type === "default" ||
+    event.type === "generic" ||
+    event.type === "csp"
+  ) {
     const typedEvent = event as
       | z.infer<typeof ErrorEventSchema>
       | z.infer<typeof DefaultEventSchema>
@@ -2200,40 +2185,17 @@ export function formatIssueOutput({
     output += `**Message**:\n${event.message}\n`;
   }
   output += "\n";
-  // only a markdown body belongs in this output; a json body is for structuredContent, and
-  // pasting it here would put a serialized object in the middle of the prose
-  if (
-    isSharedFormatterType &&
-    event.formatted?.format === "markdown" &&
-    event.formatted.content
-  ) {
-    // the shared formatter body doesn't include the replay note — add it here to match formatEventOutput
-    output += formatIssueReplayOutput({
+  output += formatEventOutput(event, {
+    performanceTrace,
+    replaySummary: {
       apiService,
       organizationSlug,
-      event,
       relatedReplayIds,
       experimentalMode: experimentalMode ?? false,
       availableToolNames,
       directToolNames,
-    });
-    const formattedContent = event.formatted.content;
-    output += formattedContent.endsWith("\n")
-      ? formattedContent
-      : `${formattedContent}\n`;
-  } else {
-    output += formatEventOutput(event, {
-      performanceTrace,
-      replaySummary: {
-        apiService,
-        organizationSlug,
-        relatedReplayIds,
-        experimentalMode: experimentalMode ?? false,
-        availableToolNames,
-        directToolNames,
-      },
-    });
-  }
+    },
+  });
 
   // Add Seer context if available
   if (autofixState) {
@@ -2249,27 +2211,67 @@ export function formatIssueOutput({
     output += "\n";
   }
 
+  output += "## Response Notes\n\n";
+  for (const note of buildIssueResponseNotes({
+    organizationSlug,
+    issue,
+    event,
+    apiService,
+    aiConversations,
+    experimentalMode,
+    availableToolNames,
+    directToolNames,
+  })) {
+    output += `- ${note}\n`;
+  }
+  return output;
+}
+
+export function buildIssueResponseNotes({
+  organizationSlug,
+  issue,
+  event,
+  apiService,
+  aiConversations,
+  experimentalMode,
+  availableToolNames,
+  directToolNames,
+}: {
+  organizationSlug: string;
+  issue: Issue;
+  event: Event;
+  apiService: SentryApiService;
+  aiConversations?: AIConversationReference[];
+  experimentalMode?: boolean;
+  availableToolNames?: ReadonlySet<string>;
+  directToolNames?: ReadonlySet<string>;
+}): string[] {
+  const notes: string[] = [];
   const traceId =
     typeof event.contexts?.trace?.trace_id === "string" &&
     event.contexts.trace.trace_id.length > 0
       ? event.contexts.trace.trace_id
       : undefined;
 
-  output += "## Response Notes\n\n";
   const commitIssueReference = /^\d+$/.test(issue.shortId)
     ? apiService.getIssueUrl(organizationSlug, issue.shortId)
     : issue.shortId;
-  output += `- Commit message issue reference: \`Fixes ${commitIssueReference}\` automatically closes the issue when the commit is merged.\n`;
-  output +=
-    "- The stacktrace includes first-party application code and third-party code. First-party frames are usually the best starting point for triage.\n";
+  notes.push(
+    `Commit message issue reference: \`Fixes ${commitIssueReference}\` automatically closes the issue when the commit is merged.`,
+  );
+  notes.push(
+    "The stacktrace includes first-party application code and third-party code. First-party frames are usually the best starting point for triage.",
+  );
   if (aiConversations && aiConversations.length > 0) {
-    output += formatAIConversationResponseNote({
-      aiConversations,
-      organizationSlug,
-      experimentalMode: experimentalMode ?? false,
-      availableToolNames,
-      directToolNames,
-    });
+    notes.push(
+      ...buildAIConversationResponseNotes({
+        aiConversations,
+        organizationSlug,
+        experimentalMode: experimentalMode ?? false,
+        availableToolNames,
+        directToolNames,
+      }),
+    );
   }
   const issueEventSearchInstruction = formatToolCallInstruction({
     toolName: "search_issue_events",
@@ -2283,7 +2285,7 @@ export function formatIssueOutput({
     directToolNames,
     fallbackInstruction: "Issue event search is not available in this session",
   });
-  output += `- Issue event search: ${issueEventSearchInstruction}\n`;
+  notes.push(`Issue event search: ${issueEventSearchInstruction}`);
   const hasMultipleThreads = event.entries?.some((entry) => {
     if (entry.type !== "threads") {
       return false;
@@ -2308,7 +2310,7 @@ export function formatIssueOutput({
         "to fetch a full thread stacktrace by numeric Thread ID or exact thread Name. Omit `thread` to use Sentry's default selected thread",
     });
     if (stacktraceInstruction) {
-      output += `- Thread stacktrace lookup: ${stacktraceInstruction}\n`;
+      notes.push(`Thread stacktrace lookup: ${stacktraceInstruction}`);
     }
   }
   if (traceId) {
@@ -2349,9 +2351,11 @@ export function formatIssueOutput({
       fallbackInstruction:
         "Related log search is not available in this session",
     });
-    output += `- Full distributed trace and span tree: ${traceDetailsInstruction}\n`;
-    output += `- Related span search: ${spanSearchInstruction}\n`;
-    output += `- Related log search: ${logSearchInstruction}\n`;
+    notes.push(
+      `Full distributed trace and span tree: ${traceDetailsInstruction}`,
+    );
+    notes.push(`Related span search: ${spanSearchInstruction}`);
+    notes.push(`Related log search: ${logSearchInstruction}`);
   }
   if (experimentalMode) {
     const breadcrumbsInstruction = formatToolCallInstruction({
@@ -2365,12 +2369,14 @@ export function formatIssueOutput({
       fallbackInstruction:
         "Issue breadcrumbs are not available in this session",
     });
-    output += `- Breadcrumb trail leading up to this error: ${breadcrumbsInstruction}\n`;
+    notes.push(
+      `Breadcrumb trail leading up to this error: ${breadcrumbsInstruction}`,
+    );
   }
-  return output;
+  return notes;
 }
 
-function formatAIConversationResponseNote({
+function buildAIConversationResponseNotes({
   aiConversations,
   organizationSlug,
   experimentalMode,
@@ -2382,7 +2388,7 @@ function formatAIConversationResponseNote({
   experimentalMode: boolean;
   availableToolNames?: ReadonlySet<string>;
   directToolNames?: ReadonlySet<string>;
-}): string {
+}): string[] {
   const instructions = formatAIConversationActionInstructions({
     organizationSlug,
     aiConversations,
@@ -2396,13 +2402,19 @@ function formatAIConversationResponseNote({
     const spanSuffix = conversation.spanId
       ? ` Matching span: \`${conversation.spanId}\`.`
       : "";
-    return `- Agent conversation found in this trace: \`${conversation.conversationId}\`.${spanSuffix}\n${instructions.map((instruction) => `- ${instruction}`).join("\n")}\n`;
+    return [
+      `Agent conversation found in this trace: \`${conversation.conversationId}\`.${spanSuffix}`,
+      ...instructions,
+    ];
   }
 
   const conversationIds = aiConversations
     .map((conversation) => `\`${conversation.conversationId}\``)
     .join(", ");
-  return `- Multiple agent conversations were found in this trace: ${conversationIds}.\n${instructions.map((instruction) => `- ${instruction}`).join("\n")}\n`;
+  return [
+    `Multiple agent conversations were found in this trace: ${conversationIds}.`,
+    ...instructions,
+  ];
 }
 
 const MAX_DISPLAY_REPLAYS = 5;
@@ -2540,7 +2552,7 @@ function normalizeReplayId(replayId: string | null | undefined): string | null {
   return trimmedReplayId.replace(/-/g, "");
 }
 
-function stripReplayMetadata(event: Event): Event {
+export function stripReplayMetadata(event: Event): Event {
   const tags = event.tags?.filter(
     (tag) => tag.key !== "replay.id" && tag.key !== "replayId",
   );

packages/mcp-core/src/internal/tool-helpers/seer.ts

@@ -116,20 +116,6 @@ function wrapSeerAnalysisOutput({
   return `<seer_analysis ${attrs.join(" ")}>\n${output.trimEnd()}\n</seer_analysis>\n`;
 }
 
-/**
- * Wraps shared-formatter Seer analysis content in the provenance boundary,
- * mirroring the tags getOutputForAutofixRun applies to MCP-rendered output.
- * Seer content is LLM-generated, so it must be marked as untrusted data.
- */
-export function wrapSeerContent(content: string, runId?: number): string {
-  return wrapSeerAnalysisOutput({
-    output: content,
-    runId,
-    step: "analysis",
-    includeProvenanceTags: true,
-  });
-}
-
 // Artifact data shapes from getsentry/sentry's
 // `src/sentry/seer/autofix/artifact_schemas.py`. Fields are LLM-generated, so
 // everything is treated as optional.