diff --git a/src/fetch/README.md b/src/fetch/README.md
index 2c3e048927..add58c2c11 100644
--- a/src/fetch/README.md
+++ b/src/fetch/README.md
@@ -24,6 +24,12 @@ The fetch tool will truncate the response, but by using the `start_index` argume
   - Arguments:
     - `url` (string, required): URL to fetch
 
+### Content negotiation
+
+The fetch tool sends an `Accept: text/markdown, text/html;q=0.9, */*;q=0.8` request header, asking servers for native markdown when they support it. If the server responds with `Content-Type: text/markdown` (with or without a charset parameter), the response body is returned as-is and the HTML-to-markdown extraction step is skipped. Otherwise the existing readability + markdownify pipeline runs as before.
+
+This benefits sites that serve markdown directly via content negotiation — for example, Cloudflare-hosted sites with the [Markdown for Agents](https://developers.cloudflare.com/fundamentals/reference/markdown-for-agents/) feature enabled (the linked documentation page is itself such a site), content-negotiating CMSes, and raw-content endpoints. Servers that don't recognise the Accept header simply respond with whatever they normally would, so the change is fully backwards-compatible.
+
 ## Installation
 
 Optionally: Install node.js, this will cause the fetch server to use a different HTML simplifier that is more robust.
diff --git a/src/fetch/src/mcp_server_fetch/server.py b/src/fetch/src/mcp_server_fetch/server.py
index b42c7b1f6b..9dac6fff48 100644
--- a/src/fetch/src/mcp_server_fetch/server.py
+++ b/src/fetch/src/mcp_server_fetch/server.py
@@ -121,7 +121,10 @@ async def fetch_url(
             response = await client.get(
                 url,
                 follow_redirects=True,
-                headers={"User-Agent": user_agent},
+                headers={
+                    "User-Agent": user_agent,
+                    "Accept": "text/markdown, text/html;q=0.9, */*;q=0.8",
+                },
                 timeout=30,
             )
         except HTTPError as e:
@@ -135,6 +138,11 @@ async def fetch_url(
         page_raw = response.text
 
     content_type = response.headers.get("content-type", "")
+
+    # Server provided markdown directly via content negotiation; skip HTML extraction.
+    if content_type.split(";", 1)[0].strip().lower() == "text/markdown":
+        return page_raw, ""
+
     is_page_html = (
         "<html" in page_raw[:100] or "text/html" in content_type or not content_type
     )
diff --git a/src/fetch/tests/test_server.py b/src/fetch/tests/test_server.py
index 96c1cb38c7..150368a6c7 100644
--- a/src/fetch/tests/test_server.py
+++ b/src/fetch/tests/test_server.py
@@ -324,3 +324,102 @@ async def test_fetch_with_proxy(self):
 
             # Verify AsyncClient was called with proxy
             mock_client_class.assert_called_once_with(proxy="http://proxy.example.com:8080")
+
+    @pytest.mark.asyncio
+    async def test_fetch_markdown_returns_early(self):
+        """Test that text/markdown responses skip HTML extraction and return body as-is."""
+        md_content = "# Hello World\n\nThis is markdown."
+        mock_response = MagicMock()
+        mock_response.status_code = 200
+        mock_response.text = md_content
+        mock_response.headers = {"content-type": "text/markdown"}
+
+        with patch("httpx.AsyncClient") as mock_client_class:
+            mock_client = AsyncMock()
+            mock_client.get = AsyncMock(return_value=mock_response)
+            mock_client_class.return_value.__aenter__ = AsyncMock(return_value=mock_client)
+            mock_client_class.return_value.__aexit__ = AsyncMock(return_value=None)
+
+            content, prefix = await fetch_url(
+                "https://example.com/readme.md",
+                DEFAULT_USER_AGENT_AUTONOMOUS
+            )
+
+            assert content == md_content
+            assert prefix == ""
+
+    @pytest.mark.asyncio
+    async def test_fetch_markdown_with_charset(self):
+        """Test that text/markdown with a charset parameter is recognised as markdown."""
+        md_content = "# Title\n\nBody."
+        mock_response = MagicMock()
+        mock_response.status_code = 200
+        mock_response.text = md_content
+        mock_response.headers = {"content-type": "text/markdown; charset=utf-8"}
+
+        with patch("httpx.AsyncClient") as mock_client_class:
+            mock_client = AsyncMock()
+            mock_client.get = AsyncMock(return_value=mock_response)
+            mock_client_class.return_value.__aenter__ = AsyncMock(return_value=mock_client)
+            mock_client_class.return_value.__aexit__ = AsyncMock(return_value=None)
+
+            content, prefix = await fetch_url(
+                "https://example.com/page.md",
+                DEFAULT_USER_AGENT_AUTONOMOUS
+            )
+
+            assert content == md_content
+            assert prefix == ""
+
+    @pytest.mark.asyncio
+    async def test_fetch_x_markdown_does_not_match(self):
+        """Test that non-standard text/x-markdown is NOT treated as markdown and falls through."""
+        body = "some non-standard markdown variant"
+        mock_response = MagicMock()
+        mock_response.status_code = 200
+        mock_response.text = body
+        mock_response.headers = {"content-type": "text/x-markdown"}
+
+        with patch("httpx.AsyncClient") as mock_client_class:
+            mock_client = AsyncMock()
+            mock_client.get = AsyncMock(return_value=mock_response)
+            mock_client_class.return_value.__aenter__ = AsyncMock(return_value=mock_client)
+            mock_client_class.return_value.__aexit__ = AsyncMock(return_value=None)
+
+            content, prefix = await fetch_url(
+                "https://example.com/page",
+                DEFAULT_USER_AGENT_AUTONOMOUS
+            )
+
+            # text/x-markdown is neither text/html nor text/markdown, so it falls
+            # into the raw-fallback branch with the "cannot be simplified" prefix.
+            # This pins the contract: only the standard text/markdown media type
+            # triggers the native-markdown short-circuit.
+            assert content == body
+            assert "cannot be simplified" in prefix
+
+    @pytest.mark.asyncio
+    async def test_fetch_sends_accept_header(self):
+        """Test that fetch_url sends an Accept header advertising markdown preference."""
+        mock_response = MagicMock()
+        mock_response.status_code = 200
+        mock_response.text = "# md"
+        mock_response.headers = {"content-type": "text/markdown"}
+
+        with patch("httpx.AsyncClient") as mock_client_class:
+            mock_client = AsyncMock()
+            mock_client.get = AsyncMock(return_value=mock_response)
+            mock_client_class.return_value.__aenter__ = AsyncMock(return_value=mock_client)
+            mock_client_class.return_value.__aexit__ = AsyncMock(return_value=None)
+
+            await fetch_url(
+                "https://example.com/page",
+                DEFAULT_USER_AGENT_AUTONOMOUS
+            )
+
+            call_kwargs = mock_client.get.call_args.kwargs
+            assert "Accept" in call_kwargs["headers"]
+            accept = call_kwargs["headers"]["Accept"]
+            assert "text/markdown" in accept
+            # Markdown should be preferred over HTML — appear earlier in the q-list.
+            assert accept.index("text/markdown") < accept.index("text/html")