[Approved | On Hold] feat: restrict access to "LangGraph Agents" provider

api/server/middleware/accessResources/canAccessAgentFromBody.js

@@ -50,7 +50,21 @@ const canAccessAgentFromBody = (options) => {
       const { endpoint, agent_id } = req.body;
       let agentId = agent_id;
 
+      // Check if endpoint is in allowedProviders (provider-only endpoints)
+      // These endpoints can only be used through created agents, not via ephemeral access
+      const allowedProviders = req.config?.endpoints?.agents?.allowedProviders || [];
+      const isProviderOnly = allowedProviders.includes(endpoint);
+
       if (!isAgentsEndpoint(endpoint)) {
+        // If this is a provider-only endpoint, completely block direct access
+        // These must go through the agents endpoint, not as standalone requests
+        if (isProviderOnly) {
+          return res.status(403).json({
+            error: 'Forbidden',
+            message:
+              'This endpoint can only be accessed through created agents with proper permissions',
+          });
+        }
         agentId = Constants.EPHEMERAL_AGENT_ID;
       }
 

librechat.n1.yml

@@ -73,6 +73,9 @@ mcpServers:
     chatMenu: false
     requiresOAuth: false
 endpoints:
+  agents:
+    allowedProviders:
+      - "LangGraph Agents"
   azureOpenAI:
     # Endpoint-level configuration
     titleModel: "gpt-4o"

librechat.n2a.yml

@@ -74,6 +74,9 @@ mcpServers:
     chatMenu: false
     requiresOAuth: false
 endpoints:
+  agents:
+    allowedProviders:
+      - "LangGraph Agents"
   azureOpenAI:
     # Endpoint-level configuration
     titleModel: "gpt-4o"

librechat.prod.yml

@@ -72,6 +72,9 @@ mcpServers:
     chatMenu: false
     requiresOAuth: false
 endpoints:
+  agents:
+    allowedProviders:
+      - "LangGraph Agents"
   azureOpenAI:
     # Endpoint-level configuration
     titleModel: "gpt-4o"