diff --git a/src/sequentialthinking/__tests__/input-schema.test.ts b/src/sequentialthinking/__tests__/input-schema.test.ts
index 4ff7be663c..472eb6b53a 100644
--- a/src/sequentialthinking/__tests__/input-schema.test.ts
+++ b/src/sequentialthinking/__tests__/input-schema.test.ts
@@ -39,6 +39,16 @@ describe.skipIf(!existsSync(distIndexPath))('sequentialthinking input schema', (
     );
   });
 
+  // Regression coverage for #4721: each call accumulates in-memory thought
+  // history, so the tool must not advertise itself as read-only or idempotent.
+  it('marks the tool as neither read-only nor idempotent', async () => {
+    const { tools } = await client.listTools();
+    const tool = tools.find(t => t.name === 'sequentialthinking');
+    expect(tool).toBeDefined();
+    expect(tool!.annotations?.readOnlyHint).toBe(false);
+    expect(tool!.annotations?.idempotentHint).toBe(false);
+  });
+
   it('rejects a call that omits nextThoughtNeeded', async () => {
     const result = await client.callTool({
       name: 'sequentialthinking',
diff --git a/src/sequentialthinking/index.ts b/src/sequentialthinking/index.ts
index 1ae09d1db8..68899ff10e 100644
--- a/src/sequentialthinking/index.ts
+++ b/src/sequentialthinking/index.ts
@@ -93,9 +93,11 @@ You should:
       needsMoreThoughts: coercedBoolean.optional().describe("If more thoughts are needed")
     },
     annotations: {
-      readOnlyHint: true,
+      // Each call appends to the server's in-memory thought history (and
+      // branch list), so the tool is neither read-only nor idempotent.
+      readOnlyHint: false,
       destructiveHint: false,
-      idempotentHint: true,
+      idempotentHint: false,
       openWorldHint: false,
     },
     outputSchema: {