diff --git a/docs/contributing/search-events-api-patterns.md b/docs/contributing/search-events-api-patterns.md
index b6299c234..5fe627927 100644
--- a/docs/contributing/search-events-api-patterns.md
+++ b/docs/contributing/search-events-api-patterns.md
@@ -114,6 +114,7 @@ https://us.sentry.io/api/0/organizations/sentry/events/?dataset=spans&field=ai.m
 - Does NOT support timestamp filters in query (use `statsPeriod` instead)
 - Severity levels: fatal, error, warning, info, debug, trace
 - Common aggregate functions: `count()`, `epm()`
+- Regex filters on string attributes: `message://timeout after \d+ms//` (RE2, unquoted, max 64 characters); prefer wildcards for plain substrings, prefixes, and suffixes
 
 #### Metrics Dataset
 - Represents newer span metrics, including counters, gauges, and distributions
diff --git a/packages/mcp-core/src/tools/support/search-events/config.ts b/packages/mcp-core/src/tools/support/search-events/config.ts
index dab182818..fc6d8299a 100644
--- a/packages/mcp-core/src/tools/support/search-events/config.ts
+++ b/packages/mcp-core/src/tools/support/search-events/config.ts
@@ -123,6 +123,21 @@ REPLAY SEARCH RULES:
 - If the user asks about replays they have viewed, prefer viewed_by_me:true
 - If the user asks about replay users and says "me", use whoami and translate to user.email:<actual email>
 
+LOGS TEXT MATCHING (LOGS DATASET ONLY):
+- Plain word, phrase, prefix, or suffix: use wildcards. message:"*database*" contains, message:"database*" starts with, message:"*database" ends with. Do NOT use a regex when a wildcard is enough
+- Any of several plain words or phrases: use a wildcard list, e.g. message:["*ConnectionReset*","*ReadTimeout*"]
+- Use a regex filter key://pattern// when wildcards cannot express the request: number shapes (\\d+), character classes ([0-9a-f]), structured values like IPs, UUIDs, and status codes, or ignoring case (wildcards are case sensitive). Use alternation (a|b) and anchors (^ or $) only alongside one of these
+  - Example: message://request took \\d+ms//
+  - Negate with a leading !: !message://worker \\d+ ready//
+  - There is no regex list form (key:[//a//,//b//] is a literal list); put alternatives inside one pattern: message://(upload|download) of \\d+ bytes failed//
+  - For an exact set of values, keep the list: severity:[error,fatal], not severity://error|fatal//
+- Regex on message is for text inside the message. When the user asks about a value that has its own attribute (e.g. response status), filter that attribute instead
+- NEVER quote a regex: message:"//...//" is a literal string match, not a regex
+- The pattern ends at the first // followed by a space, ) or the end of the query. Spaces, parentheses, and a // followed by anything else (https?://host) are fine unquoted; write \\/\\/ for a literal // followed by a space or )
+- Regex uses RE2 syntax (no lookarounds or backreferences), matches anywhere unless anchored, and is case sensitive; prefix the pattern with (?i) to ignore case
+- Patterns are limited to 64 characters (an escape like \\d counts as one), so keep them short: leave out \\b unless the request needs it, and match either case with (?i)[0-9a-f] rather than [0-9a-fA-F]
+- Regex only works on string attributes in the logs dataset. NEVER write key://pattern// for errors, spans, metrics, or any other dataset: they match //...// literally and return nothing. Approximate the shape with wildcards there, e.g. message:"*retry*failed*"
+
 MATHEMATICAL QUERY PATTERNS:
 When user asks mathematical questions like "how many X", "total Y used", "sum of Z":
 - Identify the appropriate dataset based on context
@@ -745,6 +760,33 @@ export const DATASET_EXAMPLES: Record<
         sort: "-timestamp",
       },
     },
+    {
+      description:
+        "logs whose message reports a retry count like 'retry 3 of 5'",
+      output: {
+        query: "message://retry \\d+ of \\d+//",
+        fields: ["timestamp", "message", "severity", "trace"],
+        sort: "-timestamp",
+      },
+    },
+    {
+      description:
+        "error logs whose message contains a hex error code like 'err=0x8007000e'",
+      output: {
+        query: "severity:error AND message://err=0x[0-9a-f]+//",
+        fields: ["timestamp", "message", "severity", "trace"],
+        sort: "-timestamp",
+      },
+    },
+    {
+      description:
+        "logs excluding heartbeat messages like 'heartbeat seq=1042'",
+      output: {
+        query: "!message://^heartbeat seq=\\d+$//",
+        fields: ["timestamp", "message", "severity", "trace"],
+        sort: "-timestamp",
+      },
+    },
     {
       description: "error logs excluding health checks",
       output: {
diff --git a/packages/mcp-core/src/tools/support/search-events/utils.test.ts b/packages/mcp-core/src/tools/support/search-events/utils.test.ts
index e7e84857b..b156e46f9 100644
--- a/packages/mcp-core/src/tools/support/search-events/utils.test.ts
+++ b/packages/mcp-core/src/tools/support/search-events/utils.test.ts
@@ -290,8 +290,25 @@ describe("search query helpers", () => {
     );
   });
 
+  it("should detect regex filters as structured Sentry search syntax", () => {
+    expect(
+      looksLikeSentrySearchSyntax(String.raw`message://request took \d+ms//`),
+    ).toBe(true);
+    expect(
+      looksLikeSentrySearchSyntax(String.raw`!message://worker \d+ ready//`),
+    ).toBe(true);
+    expect(
+      looksLikeSentrySearchSyntax(
+        String.raw`message://https?://api\.example\.com//`,
+      ),
+    ).toBe(true);
+  });
+
   it("should ignore common natural language colon patterns", () => {
     expect(looksLikeSentrySearchSyntax("open http://example.com")).toBe(false);
+    expect(looksLikeSentrySearchSyntax("see http://example.com/a/")).toBe(
+      false,
+    );
     expect(looksLikeSentrySearchSyntax("started at 10:30")).toBe(false);
     expect(looksLikeSentrySearchSyntax("Note: show slow spans")).toBe(false);
     expect(looksLikeSentrySearchSyntax("ERROR: service is down")).toBe(false);
@@ -392,6 +409,41 @@ describe("search query helpers", () => {
     );
     expect(isSemanticFilterDowngrade("id:1", 'message:"error 1"')).toBe(true);
   });
+
+  it("detects message regex downgrades of structured filters", () => {
+    expect(
+      isSemanticFilterDowngrade(
+        "http.status_code:500",
+        "message://status 500//",
+      ),
+    ).toBe(true);
+    expect(
+      isSemanticFilterDowngrade(
+        "user.id:42 severity:error",
+        String.raw`severity:error message://(?i)user 42 failed//`,
+      ),
+    ).toBe(true);
+
+    expect(
+      isSemanticFilterDowngrade(
+        "http.status_code:500",
+        String.raw`http.status_code:500 message://took \d+ms//`,
+      ),
+    ).toBe(false);
+
+    // A key:value inside a regex is pattern text, not a kept structured filter.
+    expect(
+      isSemanticFilterDowngrade("custom:foo", "message://custom:foo//"),
+    ).toBe(true);
+
+    // A quote inside a regex must not hide the filters after it.
+    expect(
+      isSemanticFilterDowngrade(
+        "custom:foo",
+        String.raw`message://it's \d+// message:"*foo*"`,
+      ),
+    ).toBe(true);
+  });
 });
 
 describe("fetchCustomAttributes", () => {
diff --git a/packages/mcp-core/src/tools/support/search-events/utils.ts b/packages/mcp-core/src/tools/support/search-events/utils.ts
index a8d699d7c..3e6b67daf 100644
--- a/packages/mcp-core/src/tools/support/search-events/utils.ts
+++ b/packages/mcp-core/src/tools/support/search-events/utils.ts
@@ -30,8 +30,14 @@ export type FlexibleEventData = Record<string, unknown>;
 
 const DEFAULT_MAX_VALUE_LENGTH = 200;
 const DEFAULT_MAX_ARRAY_ITEMS = 20;
-const SENTRY_SEARCH_TOKEN_PATTERN =
-  /(^|\s)!?([A-Za-z_][A-Za-z0-9_.[\],-]*):(?=\S)(?!\/\/)/g;
+// Mirrors regex_value in Sentry's event_search.py grammar, which is what
+// separates a key://pattern// filter from prose like http://example.com.
+const REGEX_FILTER_VALUE_SOURCE = String.raw`\/\/(?!\/\/(?:[\t\n )]|$))[^\n]{1,1024}?\/\/(?=[\t\n )]|$)`;
+const REGEX_FILTER_VALUE_PATTERN = new RegExp(`^${REGEX_FILTER_VALUE_SOURCE}`);
+const SENTRY_SEARCH_TOKEN_PATTERN = new RegExp(
+  String.raw`(^|\s)!?([A-Za-z_][A-Za-z0-9_.[\],-]*):(?=\S)(?:(?!\/\/)|(?=${REGEX_FILTER_VALUE_SOURCE}))`,
+  "g",
+);
 const KNOWN_SENTRY_SEARCH_KEYS = new Set([
   "browser",
   "device",
@@ -122,8 +128,9 @@ export function looksLikeSentrySearchSyntax(query?: string): boolean {
 const FULL_TEXT_SEARCH_KEYS = new Set(["message", "log.body"]);
 
 /**
- * Replace quoted regions with same-length placeholders so colons inside quotes
- * are not treated as filter-key separators (e.g. transaction:"handle message:hello").
+ * Replace quoted regions and regex filter values with same-length placeholders
+ * so colons inside them are not treated as filter-key separators
+ * (e.g. transaction:"handle message:hello" or message://key:\d+//).
  * Length is preserved so offsets still map back to the original query.
  */
 function maskQuotedRegions(query: string): string {
@@ -131,7 +138,17 @@ function maskQuotedRegions(query: string): string {
   let quote: '"' | "'" | null = null;
   let escaped = false;
 
-  for (const char of query) {
+  for (let i = 0; i < query.length; i += 1) {
+    const char = query[i] ?? "";
+    if (!quote && !escaped && char === "/" && query[i - 1] === ":") {
+      const regexValue = query.slice(i).match(REGEX_FILTER_VALUE_PATTERN)?.[0];
+      if (regexValue) {
+        // Keep the // delimiters so the token pattern still sees a regex value.
+        out += `//${regexValue.slice(2, -2).replace(/[\s:"']/g, "x")}//`;
+        i += regexValue.length - 1;
+        continue;
+      }
+    }
     if (escaped) {
       escaped = false;
       out += quote ? "x" : char;
@@ -169,6 +186,10 @@ type SearchFilterOccurrence = {
 };
 
 function normalizeFilterValue(rawValue: string): string {
+  if (REGEX_FILTER_VALUE_PATTERN.test(rawValue)) {
+    return rawValue.slice(2, -2).trim().toLowerCase();
+  }
+
   return rawValue
     .replace(/^['"]|['"]$/g, "")
     .replace(/^\*+|\*+$/g, "")
@@ -178,7 +199,8 @@ function normalizeFilterValue(rawValue: string): string {
 
 /**
  * Read one filter value starting at `valueStart` in the original query.
- * Quoted values keep interior whitespace; unquoted values stop at whitespace.
+ * Quoted and regex values keep interior whitespace; other values stop at
+ * whitespace.
  */
 function readRawFilterValue(
   query: string,
@@ -188,6 +210,13 @@ function readRawFilterValue(
     return undefined;
   }
 
+  const regexValue = query
+    .slice(valueStart)
+    .match(REGEX_FILTER_VALUE_PATTERN)?.[0];
+  if (regexValue) {
+    return regexValue;
+  }
+
   const first = query[valueStart];
   if (first === '"' || first === "'") {
     let escaped = false;