diff --git a/server/routes/mcp/index.ts b/server/routes/mcp/index.ts
index df5434e8fc..4c99f414c5 100644
--- a/server/routes/mcp/index.ts
+++ b/server/routes/mcp/index.ts
@@ -23,15 +23,21 @@ import { version } from "../../../package.json";
 const app = new Koa();
 const router = new Router();
 
+const defaultInstructions = `Document and collection markdown support @mentions using the syntax: @[Display Name](mention://user/userId). For example: @[John Doe](mention://user/c9a1b2e3-...). Use the list_users tool to find user IDs.`;
+
 /**
  * Creates a fresh MCP server instance with tools filtered by the OAuth
  * scopes granted to the current token.
  *
  * @param scopes - the OAuth scopes granted to the access token.
- * @param instructions - optional workspace guidance to send to clients.
+ * @param guidance - optional workspace guidance to append to default instructions.
  * @returns a configured McpServer ready to be connected to a transport.
  */
-function createMcpServer(scopes: string[], instructions?: string): McpServer {
+function createMcpServer(scopes: string[], guidance?: string): McpServer {
+  const instructions = guidance
+    ? `${defaultInstructions}\n\n${guidance}`
+    : defaultInstructions;
+
   const server = new McpServer(
     {
       name: "outline",
@@ -41,7 +47,7 @@ function createMcpServer(scopes: string[], instructions?: string): McpServer {
       capabilities: {
         tools: {},
       },
-      ...(instructions ? { instructions } : {}),
+      instructions,
     }
   );
 
diff --git a/server/validation.ts b/server/validation.ts
index 93166b50ab..ff8e18a2b9 100644
--- a/server/validation.ts
+++ b/server/validation.ts
@@ -251,9 +251,10 @@ export class ValidateURL {
 
       const { id, mentionType, modelId } = parseMentionUrl(url);
       return (
-        id &&
-        isUUID(id) &&
+        (!id || isUUID(id)) &&
+        !!mentionType &&
         Object.values(MentionType).includes(mentionType as MentionType) &&
+        !!modelId &&
         isUUID(modelId)
       );
     } catch (_err) {
diff --git a/shared/editor/rules/mention.test.ts b/shared/editor/rules/mention.test.ts
new file mode 100644
index 0000000000..5806668b68
--- /dev/null
+++ b/shared/editor/rules/mention.test.ts
@@ -0,0 +1,155 @@
+import markdownit from "markdown-it";
+import mentionRule from "./mention";
+
+function findMentionTokens(tokens: ReturnType<markdownit["parse"]>) {
+  const mentions: Array<{
+    id: string | null;
+    type: string | null;
+    modelId: string | null;
+    label: string;
+  }> = [];
+
+  for (const tok of tokens) {
+    if (tok.type === "inline" && tok.children) {
+      for (const child of tok.children) {
+        if (child.type === "mention") {
+          mentions.push({
+            id: child.attrGet("id"),
+            type: child.attrGet("type"),
+            modelId: child.attrGet("modelId"),
+            label: child.content,
+          });
+        }
+      }
+    }
+  }
+
+  return mentions;
+}
+
+describe("mention rule", () => {
+  const md = markdownit().use(mentionRule);
+
+  describe("3-segment format (existing)", () => {
+    it("should parse a user mention", () => {
+      const result = md.parse(
+        "@[John Doe](mention://a1b2c3d4-e5f6-7890-abcd-ef1234567890/user/f0e1d2c3-b4a5-6789-0abc-def123456789)",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0]).toEqual({
+        id: "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
+        type: "user",
+        modelId: "f0e1d2c3-b4a5-6789-0abc-def123456789",
+        label: "John Doe",
+      });
+    });
+
+    it("should parse a group mention", () => {
+      const result = md.parse(
+        "@[Engineering](mention://a1b2c3d4-e5f6-7890-abcd-ef1234567890/group/f0e1d2c3-b4a5-6789-0abc-def123456789)",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0].type).toBe("group");
+      expect(mentions[0].label).toBe("Engineering");
+    });
+  });
+
+  describe("2-segment format (new)", () => {
+    it("should parse a user mention", () => {
+      const result = md.parse(
+        "@[John Doe](mention://user/f0e1d2c3-b4a5-6789-0abc-def123456789)",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0].type).toBe("user");
+      expect(mentions[0].modelId).toBe("f0e1d2c3-b4a5-6789-0abc-def123456789");
+      expect(mentions[0].label).toBe("John Doe");
+      // instanceId should be auto-generated
+      expect(mentions[0].id).toBeTruthy();
+      expect(mentions[0].id).toMatch(
+        /^[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}$/
+      );
+    });
+
+    it("should parse a group mention", () => {
+      const result = md.parse("@[Engineering](mention://group/abc123)", {});
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0].type).toBe("group");
+      expect(mentions[0].modelId).toBe("abc123");
+      expect(mentions[0].label).toBe("Engineering");
+    });
+
+    it("should parse mention with single-word name", () => {
+      const result = md.parse("@[Alice](mention://user/abc123)", {});
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0].label).toBe("Alice");
+    });
+  });
+
+  describe("mixed content", () => {
+    it("should parse mention within text", () => {
+      const result = md.parse(
+        "Hello @[John Doe](mention://user/abc123) please review",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(1);
+      expect(mentions[0].label).toBe("John Doe");
+    });
+
+    it("should parse multiple mentions", () => {
+      const result = md.parse(
+        "@[Alice](mention://user/id1) and @[Bob](mention://user/id2)",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(2);
+      expect(mentions[0].label).toBe("Alice");
+      expect(mentions[1].label).toBe("Bob");
+    });
+
+    it("should parse mix of 2-segment and 3-segment mentions", () => {
+      const result = md.parse(
+        "@[Alice](mention://user/id1) and @[Bob](mention://inst-id/user/id2)",
+        {}
+      );
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(2);
+      expect(mentions[0].label).toBe("Alice");
+      expect(mentions[0].id).toBeTruthy(); // auto-generated
+      expect(mentions[1].label).toBe("Bob");
+      expect(mentions[1].id).toBe("inst-id");
+    });
+  });
+
+  describe("non-mentions", () => {
+    it("should not parse regular links as mentions", () => {
+      const result = md.parse("[John Doe](https://example.com)", {});
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(0);
+    });
+
+    it("should not parse links without @ prefix as mentions", () => {
+      const result = md.parse("[John Doe](mention://user/abc123)", {});
+      const mentions = findMentionTokens(result);
+
+      expect(mentions).toHaveLength(0);
+    });
+  });
+});
diff --git a/shared/editor/rules/mention.ts b/shared/editor/rules/mention.ts
index 8c5f15fa31..0f5ef0834f 100644
--- a/shared/editor/rules/mention.ts
+++ b/shared/editor/rules/mention.ts
@@ -1,5 +1,40 @@
 import type { Token, StateCore } from "markdown-it";
 import type MarkdownIt from "markdown-it";
+import { v4 as uuidv4 } from "uuid";
+import parseMentionUrl from "@shared/utils/parseMentionUrl";
+
+/**
+ * Check whether a URL is a valid mention:// href.
+ *
+ * @param href the URL string to test.
+ * @returns true when the href is a recognised mention URL.
+ */
+function isMentionHref(href: string) {
+  const { mentionType, modelId } = parseMentionUrl(href);
+  return mentionType !== undefined && modelId !== undefined;
+}
+
+/**
+ * Parse a mention:// href into the id, type and modelId needed by the editor.
+ * For 2-segment URLs (no instance id) a fresh UUID is generated.
+ *
+ * @param href the mention URL to parse.
+ * @returns the parsed components.
+ * @throws when the href is not a valid mention URL.
+ */
+function parseMentionHref(href: string): {
+  id: string;
+  type: string;
+  modelId: string;
+} {
+  const { id, mentionType, modelId } = parseMentionUrl(href);
+
+  if (!mentionType || !modelId) {
+    throw new Error(`Invalid mention href: ${href}`);
+  }
+
+  return { id: id ?? uuidv4(), type: mentionType, modelId };
+}
 
 function renderMention(tokens: Token[], idx: number) {
   const id = tokens[idx].attrGet("id");
@@ -11,8 +46,6 @@ function renderMention(tokens: Token[], idx: number) {
 }
 
 function parseMentions(state: StateCore) {
-  const hrefRE = /^mention:\/\/([a-z0-9-]+)\/([a-z]+)\/([a-z0-9-]+)$/;
-
   for (let i = 0; i < state.tokens.length; i++) {
     const tok = state.tokens[i];
     if (!(tok.type === "inline" && tok.children)) {
@@ -43,7 +76,7 @@ function parseMentions(state: StateCore) {
 
       // "link_open" token should have valid href
       const attr = openToken.attrs?.[0];
-      if (!(attr && attr[0] === "href" && hrefRE.test(attr[1]))) {
+      if (!(attr && attr[0] === "href" && isMentionHref(attr[1]))) {
         return false;
       }
 
@@ -57,11 +90,10 @@ function parseMentions(state: StateCore) {
       // remove "@" from preceding token
       precToken.content = precToken.content.slice(0, -1);
 
-      // href must be present, otherwise the hrefRE test in canChunkComposeMentionToken would've failed
+      // href must be present, otherwise the isMentionHref test would've failed
       // oxlint-disable-next-line @typescript-eslint/no-non-null-assertion
       const href = openToken.attrs![0][1];
-      const matches = href.match(hrefRE);
-      const [id, mType, mId] = matches!.slice(1);
+      const { id, type: mType, modelId: mId } = parseMentionHref(href);
 
       const mentionToken = new state.Token("mention", "", 0);
       mentionToken.attrSet("id", id);
diff --git a/shared/utils/parseMentionUrl.test.ts b/shared/utils/parseMentionUrl.test.ts
new file mode 100644
index 0000000000..0e4d896f73
--- /dev/null
+++ b/shared/utils/parseMentionUrl.test.ts
@@ -0,0 +1,58 @@
+import parseMentionUrl from "./parseMentionUrl";
+
+describe("parseMentionUrl", () => {
+  it("should parse 3-segment mention URL", () => {
+    expect(
+      parseMentionUrl(
+        "mention://9a17c1c8-d178-4350-9001-203a73070fcb/user/abc123def456"
+      )
+    ).toEqual({
+      id: "9a17c1c8-d178-4350-9001-203a73070fcb",
+      mentionType: "user",
+      modelId: "abc123def456",
+    });
+  });
+
+  it("should parse 2-segment mention URL", () => {
+    expect(parseMentionUrl("mention://user/abc123def456")).toEqual({
+      mentionType: "user",
+      modelId: "abc123def456",
+    });
+  });
+
+  it("should parse 2-segment mention URL with UUID modelId", () => {
+    expect(
+      parseMentionUrl("mention://user/9a17c1c8-d178-4350-9001-203a73070fcb")
+    ).toEqual({
+      mentionType: "user",
+      modelId: "9a17c1c8-d178-4350-9001-203a73070fcb",
+    });
+  });
+
+  it("should parse group mention type", () => {
+    expect(parseMentionUrl("mention://group/abc123")).toEqual({
+      mentionType: "group",
+      modelId: "abc123",
+    });
+  });
+
+  it("should parse pull_request mention type with underscore", () => {
+    expect(
+      parseMentionUrl(
+        "mention://9a17c1c8-d178-4350-9001-203a73070fcb/pull_request/abc123"
+      )
+    ).toEqual({
+      id: "9a17c1c8-d178-4350-9001-203a73070fcb",
+      mentionType: "pull_request",
+      modelId: "abc123",
+    });
+  });
+
+  it("should return empty object for invalid URL", () => {
+    expect(parseMentionUrl("https://example.com")).toEqual({});
+  });
+
+  it("should return empty object for empty string", () => {
+    expect(parseMentionUrl("")).toEqual({});
+  });
+});
diff --git a/shared/utils/parseMentionUrl.ts b/shared/utils/parseMentionUrl.ts
index f4ab806074..f746f93dc4 100644
--- a/shared/utils/parseMentionUrl.ts
+++ b/shared/utils/parseMentionUrl.ts
@@ -1,12 +1,30 @@
-const parseMentionUrl = (url: string) => {
-  const matches = url.match(
-    /^mention:\/\/([a-z0-9-]+)\/([a-z]+)\/([a-z0-9-]+)$/
+/**
+ * Parse a mention:// URL into its components.
+ *
+ * Supports both the 3-segment format (mention://id/type/modelId) and the
+ * 2-segment format (mention://type/modelId).
+ *
+ * @param url the mention URL to parse.
+ * @returns the parsed components, or an empty object if the URL is invalid.
+ */
+const parseMentionUrl = (
+  url: string
+): { id?: string; mentionType?: string; modelId?: string } => {
+  const match3 = url.match(
+    /^mention:\/\/([a-z0-9-]+)\/([a-z_]+)\/([a-z0-9-]+)$/
   );
-  if (!matches) {
-    return {};
+  if (match3) {
+    const [id, mentionType, modelId] = match3.slice(1);
+    return { id, mentionType, modelId };
   }
-  const [id, mentionType, modelId] = matches.slice(1);
-  return { id, mentionType, modelId };
+
+  const match2 = url.match(/^mention:\/\/([a-z_]+)\/([a-z0-9-]+)$/);
+  if (match2) {
+    const [mentionType, modelId] = match2.slice(1);
+    return { mentionType, modelId };
+  }
+
+  return {};
 };
 
 export default parseMentionUrl;