diff --git a/.changeset/stdio-byte-order-mark.md b/.changeset/stdio-byte-order-mark.md
new file mode 100644
index 0000000000..dd5f1a7d15
--- /dev/null
+++ b/.changeset/stdio-byte-order-mark.md
@@ -0,0 +1,7 @@
+---
+'@modelcontextprotocol/core-internal': patch
+'@modelcontextprotocol/client': patch
+'@modelcontextprotocol/server': patch
+---
+
+Parse stdio messages that start with a UTF-8 byte order mark. `ReadBuffer` skips lines that fail to parse as JSON, so when a peer wrote a BOM before a message (as some Windows tools and shell redirections do), that message was dropped without an error and the pending request waited for its timeout. A leading U+FEFF is now stripped from each line before parsing, which RFC 8259 §8.1 allows.
diff --git a/packages/client/test/client/stdio.test.ts b/packages/client/test/client/stdio.test.ts
index 315b8a2595..82f8dc7b88 100644
--- a/packages/client/test/client/stdio.test.ts
+++ b/packages/client/test/client/stdio.test.ts
@@ -122,6 +122,31 @@ test('should fire onerror and close when ReadBuffer overflows', async () => {
     await closed;
 });
 
+test('should read a message the server prefixes with a UTF-8 byte order mark', async () => {
+    const expected: JSONRPCMessage = { jsonrpc: '2.0', method: 'notifications/initialized' };
+    const line = JSON.stringify(expected) + '\n';
+    const client = new StdioClientTransport({
+        command: 'node',
+        args: ['-e', `process.stdout.write(Buffer.concat([Buffer.from([0xef, 0xbb, 0xbf]), Buffer.from(${JSON.stringify(line)})]))`]
+    });
+    client.onerror = error => {
+        throw error;
+    };
+
+    const readMessages: JSONRPCMessage[] = [];
+    client.onmessage = message => {
+        readMessages.push(message);
+    };
+    const closed = new Promise<void>(resolve => {
+        client.onclose = () => resolve();
+    });
+
+    await client.start();
+    await closed;
+
+    expect(readMessages).toEqual([expected]);
+});
+
 test('_dispose releases the parent-side pipe handles even when a helper process holds the child stdio', async () => {
     // The rmcp-holding anatomy: the child exits, but a helper it spawned with
     // stdio: 'inherit' keeps the pipe write ends open. Awaiting 'exit' settles
diff --git a/packages/core-internal/src/shared/stdio.ts b/packages/core-internal/src/shared/stdio.ts
index 8bd794b87b..04d6a6d680 100644
--- a/packages/core-internal/src/shared/stdio.ts
+++ b/packages/core-internal/src/shared/stdio.ts
@@ -30,7 +30,12 @@ export class ReadBuffer {
                 return null;
             }
 
-            const line = this._buffer.toString('utf8', 0, index).replace(/\r$/, '');
+            // A UTF-8 byte order mark (written by some Windows tools and shell redirections)
+            // is not part of the JSON text; RFC 8259 §8.1 lets parsers ignore it.
+            const line = this._buffer
+                .toString('utf8', 0, index)
+                .replace(/\r$/, '')
+                .replace(/^\uFEFF/, '');
             this._buffer = this._buffer.subarray(index + 1);
 
             try {
diff --git a/packages/core-internal/test/shared/stdio.test.ts b/packages/core-internal/test/shared/stdio.test.ts
index f8d27a4c1f..e9c3b0ef1c 100644
--- a/packages/core-internal/test/shared/stdio.test.ts
+++ b/packages/core-internal/test/shared/stdio.test.ts
@@ -114,6 +114,34 @@ describe('non-JSON line filtering', () => {
     });
 });
 
+describe('byte order mark', () => {
+    const utf8ByteOrderMark = Buffer.from([0xef, 0xbb, 0xbf]);
+
+    test('should parse a message preceded by a UTF-8 byte order mark', () => {
+        const readBuffer = new ReadBuffer();
+        readBuffer.append(Buffer.concat([utf8ByteOrderMark, Buffer.from(JSON.stringify(testMessage) + '\n')]));
+
+        expect(readBuffer.readMessage()).toEqual(testMessage);
+        expect(readBuffer.readMessage()).toBeNull();
+    });
+
+    test('should parse a CRLF-terminated message whose byte order mark is split across chunks', () => {
+        const readBuffer = new ReadBuffer();
+        readBuffer.append(utf8ByteOrderMark.subarray(0, 1));
+        readBuffer.append(Buffer.concat([utf8ByteOrderMark.subarray(1), Buffer.from(JSON.stringify(testMessage) + '\r\n')]));
+
+        expect(readBuffer.readMessage()).toEqual(testMessage);
+    });
+
+    test('should keep a U+FEFF character that is part of the message', () => {
+        const readBuffer = new ReadBuffer();
+        const message: JSONRPCMessage = { jsonrpc: '2.0', method: 'test', params: { text: String.fromCharCode(0xfeff) + 'hello' } };
+        readBuffer.append(Buffer.from(JSON.stringify(message) + '\n'));
+
+        expect(readBuffer.readMessage()).toEqual(message);
+    });
+});
+
 describe('buffer size limit', () => {
     test('should throw when buffer exceeds default max size', () => {
         const readBuffer = new ReadBuffer();