maxisbey · Aug 21, 2026 · Aug 14, 2026 · Aug 20, 2026
diff --git a/docs/server.md b/docs/server.md
@@ -1253,7 +1253,7 @@ The FastMCP server instance accessible via `ctx.fastmcp` provides access to serv
   - `host` and `port` - Server network configuration
   - `mount_path`, `sse_path`, `streamable_http_path` - Transport paths
   - `stateless_http` - Whether the server operates in stateless mode
-  - `max_request_body_size` - Maximum Streamable HTTP POST body size in bytes
+  - `max_request_body_size` - Maximum HTTP request body size in bytes (Streamable HTTP and SSE)
   - And other configuration options
 
 ```python
@@ -1418,9 +1418,9 @@ Note that `uv run mcp run` or `uv run mcp dev` only supports server using FastMC
 
 > **Note**: Streamable HTTP transport is the recommended transport for production deployments. Use `stateless_http=True` and `json_response=True` for optimal scalability.
 
-Streamable HTTP POST bodies are limited to 4 MiB by default. Larger requests receive HTTP 413
-before parsing or session creation. If your server intentionally accepts larger MCP messages,
-configure the smallest suitable byte limit:
+HTTP request bodies (Streamable HTTP and SSE) are limited to 4 MiB by default. Larger requests
+receive HTTP 413 before parsing or session creation. If your server intentionally accepts larger MCP
+messages, configure the smallest suitable byte limit:
 
 ```python
 mcp = FastMCP("Large messages", max_request_body_size=8 * 1024 * 1024)

diff --git a/src/mcp/server/auth/routes.py b/src/mcp/server/auth/routes.py
@@ -18,6 +18,7 @@
 from mcp.server.auth.provider import OAuthAuthorizationServerProvider
 from mcp.server.auth.settings import ClientRegistrationOptions, RevocationOptions
 from mcp.server.streamable_http import MCP_PROTOCOL_VERSION_HEADER
+from mcp.server.transport_security import DEFAULT_MAX_REQUEST_BODY_SIZE, RequestBodyLimitMiddleware
 from mcp.shared.auth import OAuthMetadata
 
 
@@ -53,17 +54,24 @@ def validate_issuer_url(url: AnyHttpUrl):
 REVOCATION_PATH = "/revoke"
 
 
-def cors_middleware(
-    handler: Callable[[Request], Response | Awaitable[Response]],
-    allow_methods: list[str],
-) -> ASGIApp:
-    cors_app = CORSMiddleware(
-        app=request_response(handler),
+def _cors(app: ASGIApp, allow_methods: list[str]) -> ASGIApp:
+    return CORSMiddleware(
+        app=app,
         allow_origins="*",
         allow_methods=allow_methods,
         allow_headers=[MCP_PROTOCOL_VERSION_HEADER],
     )
-    return cors_app
+
+
+def _body_limited(app: ASGIApp) -> ASGIApp:
+    return RequestBodyLimitMiddleware(app, DEFAULT_MAX_REQUEST_BODY_SIZE)
+
+
+def cors_middleware(
+    handler: Callable[[Request], Response | Awaitable[Response]],
+    allow_methods: list[str],
+) -> ASGIApp:
+    return _cors(request_response(handler), allow_methods)
 
 
 def create_auth_routes(
@@ -84,11 +92,13 @@ def create_auth_routes(
         revocation_options,
     )
     client_authenticator = ClientAuthenticator(provider)
+    token_handler = TokenHandler(provider, client_authenticator)
 
     # Create routes
     # Allow CORS requests for endpoints meant to be hit by the OAuth client
     # (with the client secret). This is intended to support things like MCP Inspector,
-    # where the client runs in a web browser.
+    # where the client runs in a web browser. CORS is the outermost wrapper so that
+    # responses produced by inner layers (such as a 413) still carry CORS headers.
     routes = [
         Route(
             "/.well-known/oauth-authorization-server",
@@ -102,15 +112,12 @@ def create_auth_routes(
             AUTHORIZATION_PATH,
             # do not allow CORS for authorization endpoint;
             # clients should just redirect to this
-            endpoint=AuthorizationHandler(provider).handle,
+            endpoint=_body_limited(request_response(AuthorizationHandler(provider).handle)),
             methods=["GET", "POST"],
         ),
         Route(
             TOKEN_PATH,
-            endpoint=cors_middleware(
-                TokenHandler(provider, client_authenticator).handle,
-                ["POST", "OPTIONS"],
-            ),
+            endpoint=_cors(_body_limited(request_response(token_handler.handle)), ["POST", "OPTIONS"]),
             methods=["POST", "OPTIONS"],
         ),
     ]
@@ -123,10 +130,7 @@ def create_auth_routes(
         routes.append(
             Route(
                 REGISTRATION_PATH,
-                endpoint=cors_middleware(
-                    registration_handler.handle,
-                    ["POST", "OPTIONS"],
-                ),
+                endpoint=_cors(_body_limited(request_response(registration_handler.handle)), ["POST", "OPTIONS"]),
                 methods=["POST", "OPTIONS"],
             )
         )
@@ -136,10 +140,7 @@ def create_auth_routes(
         routes.append(
             Route(
                 REVOCATION_PATH,
-                endpoint=cors_middleware(
-                    revocation_handler.handle,
-                    ["POST", "OPTIONS"],
-                ),
+                endpoint=_cors(_body_limited(request_response(revocation_handler.handle)), ["POST", "OPTIONS"]),
                 methods=["POST", "OPTIONS"],
             )
         )

diff --git a/src/mcp/server/fastmcp/server.py b/src/mcp/server/fastmcp/server.py
@@ -62,8 +62,8 @@
 from mcp.server.sse import SseServerTransport
 from mcp.server.stdio import stdio_server
 from mcp.server.streamable_http import EventStore
-from mcp.server.streamable_http_manager import DEFAULT_MAX_REQUEST_BODY_SIZE, StreamableHTTPSessionManager
-from mcp.server.transport_security import TransportSecuritySettings
+from mcp.server.streamable_http_manager import StreamableHTTPSessionManager
+from mcp.server.transport_security import DEFAULT_MAX_REQUEST_BODY_SIZE, TransportSecuritySettings
 from mcp.shared.context import LifespanContextT, RequestContext, RequestT
 from mcp.types import Annotations, AnyFunction, ContentBlock, GetPromptResult, Icon, ToolAnnotations
 from mcp.types import Prompt as MCPPrompt
@@ -107,6 +107,7 @@ class Settings(BaseSettings, Generic[LifespanResultT]):
     stateless_http: bool
     """Define if the server should create a new transport per request."""
     max_request_body_size: int
+    """Maximum request body size in bytes for the Streamable HTTP endpoint and the SSE message endpoint."""
 
     # resource settings
     warn_on_duplicate_resources: bool
@@ -835,6 +836,7 @@ def sse_app(self, mount_path: str | None = None) -> Starlette:
         sse = SseServerTransport(
             normalized_message_endpoint,
             security_settings=self.settings.transport_security,
+            max_request_body_size=self.settings.max_request_body_size,
         )
 
         async def handle_sse(scope: Scope, receive: Receive, send: Send):  # pragma: no cover

diff --git a/src/mcp/server/sse.py b/src/mcp/server/sse.py
@@ -54,6 +54,8 @@
 import mcp.types as types
 from mcp.server.auth.middleware.bearer_auth import AuthenticatedUser, AuthorizationContext, authorization_context
 from mcp.server.transport_security import (
+    DEFAULT_MAX_REQUEST_BODY_SIZE,
+    RequestBodyLimitMiddleware,
     TransportSecurityMiddleware,
     TransportSecuritySettings,
 )
@@ -81,7 +83,12 @@
     _session_owners: dict[UUID, AuthorizationContext]
     _security: TransportSecurityMiddleware
 
-    def __init__(self, endpoint: str, security_settings: TransportSecuritySettings | None = None) -> None:
+    def __init__(
+        self,
+        endpoint: str,
+        security_settings: TransportSecuritySettings | None = None,
+        max_request_body_size: int = DEFAULT_MAX_REQUEST_BODY_SIZE,
+    ) -> None:
         """
         Creates a new SSE server transport, which will direct the client to POST
         messages to the relative path given.
@@ -90,6 +97,9 @@
             endpoint: A relative path where messages should be posted
                     (e.g., "/messages/").
             security_settings: Optional security settings for DNS rebinding protection.
+            max_request_body_size: Maximum size in bytes for POSTed message bodies. Requests that
+                declare or stream a larger body receive HTTP 413. Defaults to 4 MiB, matching
+                `StreamableHTTPSessionManager`.
 
         Note:
             We use relative paths instead of full URLs for several reasons:
@@ -106,6 +116,9 @@
 
         super().__init__()
 
+        if max_request_body_size <= 0:
+            raise ValueError("max_request_body_size must be a positive number of bytes")
+
         # Validate that endpoint is a relative path and not a full URL
         if "://" in endpoint or endpoint.startswith("//") or "?" in endpoint or "#" in endpoint:
             raise ValueError(
@@ -121,6 +134,7 @@
         self._read_stream_writers = {}
         self._session_owners = {}
         self._security = TransportSecurityMiddleware(security_settings)
+        self._post_message_app = RequestBodyLimitMiddleware(self._handle_post_message, max_request_body_size)
         logger.debug(f"SseServerTransport initialized with endpoint: {endpoint}")
 
     @asynccontextmanager
@@ -214,7 +228,18 @@
             self._read_stream_writers.pop(session_id, None)
             self._session_owners.pop(session_id, None)
 
-    async def handle_post_message(self, scope: Scope, receive: Receive, send: Send) -> None:  # pragma: no cover
+    async def handle_post_message(self, scope: Scope, receive: Receive, send: Send) -> None:
+        """ASGI application for the message endpoint.
+
+        Only POST is accepted (other methods get 405), and bodies larger than
+        `max_request_body_size` are answered with 413 before the message is handled.
+        """
+        if scope["method"] != "POST":
+            response = Response(status_code=405, headers={"Allow": "POST"})
+            return await response(scope, receive, send)
+        await self._post_message_app(scope, receive, send)
+
+    async def _handle_post_message(self, scope: Scope, receive: Receive, send: Send) -> None:  # pragma: no cover
         logger.debug("Handling POST message")
         request = Request(scope, receive)
 

diff --git a/src/mcp/server/streamable_http_manager.py b/src/mcp/server/streamable_http_manager.py
@@ -4,17 +4,15 @@
 
 import contextlib
 import logging
-from collections import deque
 from collections.abc import AsyncIterator
-from typing import Any, Final
+from typing import Any
 from uuid import uuid4
 
 import anyio
 from anyio.abc import TaskStatus
-from starlette.datastructures import Headers
 from starlette.requests import Request
 from starlette.responses import Response
-from starlette.types import ASGIApp, Message, Receive, Scope, Send
+from starlette.types import Receive, Scope, Send
 
 from mcp.server.auth.middleware.bearer_auth import AuthenticatedUser, AuthorizationContext, authorization_context
 from mcp.server.lowlevel.server import Server as MCPServer
@@ -23,13 +21,16 @@
     EventStore,
     StreamableHTTPServerTransport,
 )
-from mcp.server.transport_security import TransportSecuritySettings
+from mcp.server.transport_security import (
+    DEFAULT_MAX_REQUEST_BODY_SIZE,
+    RequestBodyLimitMiddleware,
+    TransportSecuritySettings,
+)
 from mcp.types import INVALID_REQUEST, ErrorData, JSONRPCError
 
-logger = logging.getLogger(__name__)
+__all__ = ["DEFAULT_MAX_REQUEST_BODY_SIZE", "RequestBodyLimitMiddleware", "StreamableHTTPSessionManager"]
 
-DEFAULT_MAX_REQUEST_BODY_SIZE: Final = 4 * 1024 * 1024
-"""Default maximum Streamable HTTP request body size in bytes (4 MiB)."""
+logger = logging.getLogger(__name__)
 
 
 class StreamableHTTPSessionManager:
@@ -65,7 +66,7 @@ class StreamableHTTPSessionManager:
             retry_interval is also configured, ensure the idle timeout comfortably exceeds the retry interval to
             avoid reaping sessions during normal SSE polling gaps. Default is None (no timeout). A value of 1800
             (30 minutes) is recommended for most deployments.
-        max_request_body_size: Maximum size in bytes for Streamable HTTP POST request bodies. Requests that
+        max_request_body_size: Maximum size in bytes for Streamable HTTP request bodies. Requests that
             exceed this limit receive a 413 response before parsing or session creation. Defaults to 4 MiB.
     """
 
@@ -361,63 +362,3 @@ async def run_server(*, task_status: TaskStatus[None] = anyio.TASK_STATUS_IGNORE
                 body.model_dump_json(by_alias=True, exclude_none=True), status_code=404, media_type="application/json"
             )
             await response(scope, receive, send)
-
-
-class RequestBodyLimitMiddleware:
-    """Reject oversized HTTP request bodies before invoking an ASGI application."""
-
-    def __init__(self, app: ASGIApp, max_body_size: int) -> None:
-        self.app = app
-        self.max_body_size = max_body_size
-
-    async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
-        if scope["type"] != "http" or scope["method"] != "POST":
-            await self.app(scope, receive, send)
-            return
-
-        headers = Headers(scope=scope)
-        content_length = headers.get("content-length")
-        if content_length is not None:
-            try:
-                declared_size = int(content_length)
-            except ValueError:
-                pass
-            else:
-                if declared_size > self.max_body_size:
-                    response = Response("Request body too large", status_code=413)
-                    return await response(scope, receive, send)
-
-        received_body = bytearray()
-        received_request = False
-        body_complete = False
-        trailing_message: Message | None = None
-        while True:
-            message = await receive()
-            if message["type"] != "http.request":
-                trailing_message = message
-                break
-
-            received_request = True
-            body = message.get("body", b"")
-            if len(received_body) + len(body) > self.max_body_size:
-                response = Response("Request body too large", status_code=413)
-                return await response(scope, receive, send)
-            received_body.extend(body)
-            if not message.get("more_body", False):
-                body_complete = True
-                break
-
-        cached_messages: deque[Message] = deque()
-        if received_request:
-            cached_messages.append(
-                {"type": "http.request", "body": bytes(received_body), "more_body": not body_complete}
-            )
-        if trailing_message is not None:
-            cached_messages.append(trailing_message)
-
-        async def replay() -> Message:
-            if cached_messages:
-                return cached_messages.popleft()
-            return await receive()
-
-        await self.app(scope, replay, send)