fix: new MCP server installation and logs UX (#2549)

Co-authored-by: brojd <[email protected]>

Author: joeyorlando

.github/values-ci.yaml

@@ -5,6 +5,11 @@ archestra:
   # Image will be set via --set during deployment
   # image: archestra/platform:ci-test
 
+  # Deploy multiple replicas for the e2e environment to test issues we may face
+  # which could be caused by storing state in pod memory, which may only be discovered
+  # when replicas is > 1
+  replicaCount: 5
+
   # Environment variables for e2e testing
   env:
     ARCHESTRA_ANALYTICS: disabled

platform/backend/src/mcp-server-runtime/manager.test.ts

@@ -1,4 +1,5 @@
 import * as fs from "node:fs";
+import { PassThrough } from "node:stream";
 import * as k8s from "@kubernetes/client-node";
 import { vi } from "vitest";
 import type * as originalConfigModule from "@/config";
@@ -74,7 +75,11 @@ vi.mock("@/models/mcp-server", () => ({
 }));
 
 vi.mock("./k8s-deployment", () => ({
-  default: vi.fn(),
+  default: class MockK8sDeployment {
+    static sanitizeLabelValue(value: string): string {
+      return value;
+    }
+  },
 }));
 
 describe("validateKubeconfig", () => {
@@ -428,4 +433,38 @@ describe("McpServerRuntimeManager", () => {
       mockMakeApiClient.mockRestore();
     });
   });
+
+  describe("streamMcpServerLogs", () => {
+    beforeEach(() => {
+      vi.clearAllMocks();
+      vi.resetModules();
+    });
+
+    test("writes a helpful message when runtime is not configured", async () => {
+      const mockLoadFromDefault = vi
+        .spyOn(k8s.KubeConfig.prototype, "loadFromDefault")
+        .mockImplementation(() => {
+          throw new Error("Config load failed");
+        });
+
+      const { McpServerRuntimeManager } = await import("./manager");
+      const manager = new McpServerRuntimeManager();
+
+      const stream = new PassThrough();
+      let output = "";
+      stream.on("data", (chunk) => {
+        output += chunk.toString();
+      });
+
+      await manager.streamMcpServerLogs("test-server-id", stream);
+
+      expect(output).toContain("Unable to stream logs");
+      expect(output).toContain(
+        "Kubernetes runtime is not configured on this instance.",
+      );
+      expect(output).toContain("mcp-server-id=test-server-id");
+
+      mockLoadFromDefault.mockRestore();
+    });
+  });
 });

platform/backend/src/mcp-server-runtime/manager.ts

@@ -561,7 +561,8 @@ export class McpServerRuntimeManager {
     // Try to get from memory first, or lazy-load from database
     const k8sDeployment = await this.getOrLoadDeployment(mcpServerId);
     if (!k8sDeployment) {
-      throw new Error(`MCP server not found`);
+      this.writeLogsUnavailableMessage(responseStream, mcpServerId);
+      return;
     }
 
     await k8sDeployment.streamLogs(responseStream, lines, abortSignal);
@@ -658,6 +659,30 @@ export class McpServerRuntimeManager {
     await Promise.allSettled(stopPromises);
     logger.info("MCP Server Runtime shutdown complete");
   }
+
+  private writeLogsUnavailableMessage(
+    responseStream: NodeJS.WritableStream,
+    mcpServerId: string,
+  ): void {
+    if ("destroyed" in responseStream && responseStream.destroyed) {
+      return;
+    }
+
+    const reason = this.k8sApi
+      ? "Deployment not loaded in runtime."
+      : "Kubernetes runtime is not configured on this instance.";
+    const command = this.getMcpServerDescribeCommand(mcpServerId);
+    const message = [
+      "Unable to stream logs for this MCP server.",
+      reason,
+      "Try running:",
+      command,
+      "",
+    ].join("\n");
+
+    responseStream.write(message);
+    responseStream.end();
+  }
 }
 
 export default new McpServerRuntimeManager();

platform/backend/src/routes/internal-mcp-catalog.ts

@@ -453,12 +453,22 @@ const internalMcpCatalogRoutes: FastifyPluginAsyncZod = async (fastify) => {
             try {
               for (const server of installedServers) {
                 try {
+                  await McpServerModel.update(server.id, {
+                    localInstallationStatus: "pending",
+                    localInstallationError: null,
+                  });
                   await autoReinstallServer(server, catalogItem);
+                  await McpServerModel.update(server.id, {
+                    localInstallationStatus: "success",
+                    localInstallationError: null,
+                  });
                   logger.info(
                     { serverId: server.id, serverName: server.name },
                     "Auto-reinstalled MCP server successfully",
                   );
                 } catch (error) {
+                  const errorMessage =
+                    error instanceof Error ? error.message : "Unknown error";
                   logger.error(
                     {
                       err: error,
@@ -470,6 +480,8 @@ const internalMcpCatalogRoutes: FastifyPluginAsyncZod = async (fastify) => {
                   // Mark for manual reinstall on failure
                   await McpServerModel.update(server.id, {
                     reinstallRequired: true,
+                    localInstallationStatus: "error",
+                    localInstallationError: errorMessage,
                   });
                 }
               }

platform/backend/src/test/fixtures.ts

@@ -363,7 +363,7 @@ async function makeMember(
  */
 async function makeMcpServer(
   overrides: Partial<
-    Pick<InsertMcpServer, "name" | "catalogId" | "ownerId">
+    Pick<InsertMcpServer, "name" | "catalogId" | "ownerId" | "teamId">
   > = {},
 ) {
   // Create a catalog if catalogId is not provided