mcp: throttle transport read requests

apps/mcp-server/src/app.rs

@@ -179,6 +179,10 @@ async fn mcp_get(
         return status.into_response();
     }
 
+    if let Err(rejection) = enforce_transport_rate_limit(&state, &path, &headers) {
+        return rate_limited_status_response(rejection);
+    }
+
     let session_id = match session_id_from_headers(&headers) {
         Ok(Some(session_id)) => session_id,
         Ok(None) => return StatusCode::BAD_REQUEST.into_response(),
@@ -223,6 +227,10 @@ async fn mcp_delete(
         return status.into_response();
     }
 
+    if let Err(rejection) = enforce_transport_rate_limit(&state, &path, &headers) {
+        return rate_limited_status_response(rejection);
+    }
+
     match session_id_from_headers(&headers) {
         Ok(Some(session_id)) => match state.sessions.get(&session_id).await {
             Ok(Some(session))
@@ -1832,6 +1840,14 @@ fn enforce_post_rate_limit(
     state: &Arc<AppState>,
     path: &AgentRoutePath,
     headers: &HeaderMap,
+) -> Result<(), RateLimitRejection> {
+    enforce_transport_rate_limit(state, path, headers)
+}
+
+fn enforce_transport_rate_limit(
+    state: &Arc<AppState>,
+    path: &AgentRoutePath,
+    headers: &HeaderMap,
 ) -> Result<(), RateLimitRejection> {
     let key = rate_limit_key(path, headers);
     state.api_rate_limiter.check(&key)
@@ -1885,6 +1901,15 @@ fn rate_limited_jsonrpc_response(
     response
 }
 
+fn rate_limited_status_response(rejection: RateLimitRejection) -> Response {
+    let mut response = StatusCode::TOO_MANY_REQUESTS.into_response();
+    let retry_after_seconds = rejection.retry_after_ms.div_ceil(1000);
+    if let Ok(value) = HeaderValue::from_str(&retry_after_seconds.to_string()) {
+        response.headers_mut().insert(RETRY_AFTER, value);
+    }
+    response
+}
+
 fn allows_scope(scopes: &[PlatformApiKeyScope], required_scope: PlatformApiKeyScope) -> bool {
     match required_scope {
         PlatformApiKeyScope::Read => scopes.iter().any(|scope| {