diff --git a/packages/vscode-extension/package.json b/packages/vscode-extension/package.json
index 417d1dbc4..3fd4bb476 100644
--- a/packages/vscode-extension/package.json
+++ b/packages/vscode-extension/package.json
@@ -288,6 +288,18 @@
         "command": "RNIDE.removeLicense",
         "title": "Remove license",
         "category": "Radon IDE"
+      },
+      {
+        "enablement": "!RNIDE.MCPToolTestsRunning",
+        "command": "RNIDE.testChatToolUsage",
+        "title": "Test AI tool usage",
+        "category": "Radon IDE"
+      },
+      {
+        "enablement": "RNIDE.MCPToolTestsRunning",
+        "command": "RNIDE.terminateChatToolTest",
+        "title": "Terminate MCP tool tests",
+        "category": "Radon IDE"
       }
     ],
     "keybindings": [
diff --git a/packages/vscode-extension/src/ai/tests/aiChatTester.ts b/packages/vscode-extension/src/ai/tests/aiChatTester.ts
new file mode 100644
index 000000000..dfa64047a
--- /dev/null
+++ b/packages/vscode-extension/src/ai/tests/aiChatTester.ts
@@ -0,0 +1,231 @@
+import { randomBytes } from "crypto";
+import { readFileSync } from "fs";
+import { mkdtemp, rm } from "fs/promises";
+import { tmpdir } from "os";
+import path from "path";
+import { window, commands, Uri, workspace, StatusBarAlignment, ThemeColor } from "vscode";
+import { Logger } from "../../Logger";
+import { exec } from "../../utilities/subprocess";
+import { Platform } from "../../utilities/platform";
+import { IDE } from "../../project/ide";
+import { testCases } from "./chatTestCases";
+import { Response, ToolCallResponse, ChatTestResult, ChatTestCase, ChatData } from "./models";
+
+export const GIT_PATH = Platform.select({
+  macos: "git",
+  windows: "git.exe",
+  linux: "git",
+});
+
+function isToolCallResponse(response: Response): response is ToolCallResponse {
+  // Smart-casting with `Exclude<string, "literal">` does not work, which is why this utility function is necessary
+  return response.kind === "toolInvocationSerialized";
+}
+
+async function clearEdits() {
+  // Stop previous response - prevents pop-ups on `workbench.action.chat.newChat`.
+  await commands.executeCommand("workbench.action.chat.cancel");
+
+  // Move cursor to input - REQUIRED for `chatEditing.acceptAllFiles`.
+  await commands.executeCommand("workbench.panel.chat.view.copilot.focus");
+
+  // Rejection requires user confirmation, acceptance does not.
+  await commands.executeCommand("chatEditing.acceptAllFiles");
+
+  const gitUri = workspace.workspaceFolders?.[0].uri;
+
+  if (!gitUri) {
+    // This case should never occur when a test app is loaded.
+    return;
+  }
+
+  // Revert all changes via git - we CANNOT use `commands.executeCommand`, as it requires user confirmation.
+  await exec(GIT_PATH, ["-C", gitUri.fsPath, "restore", "."]);
+}
+
+async function setGlobalTestsRunning(areTestsRunning: boolean) {
+  await commands.executeCommand("setContext", "RNIDE.MCPToolTestsRunning", areTestsRunning);
+}
+
+function awaitTestTerminationOrTimeout(ideInstance: IDE, testTimeout: number): Promise<boolean> {
+  return new Promise((resolve) => {
+    const disposable = ideInstance.onStateChanged(() => {
+      // Using partial state here is much more cumbersome and less readable.
+      ideInstance.getState().then((state) => {
+        const testsRunning = state.workspaceConfiguration.radonAI.areMCPTestsRunning;
+        if (testsRunning === false) {
+          disposable.dispose();
+          clearTimeout(timeout);
+          resolve(false);
+        }
+      });
+    });
+
+    const timeout = setTimeout(() => {
+      disposable.dispose();
+      resolve(true);
+    }, testTimeout);
+  });
+}
+
+async function setTestStatus(areTestsRunning: boolean, ideInstance: IDE) {
+  await setGlobalTestsRunning(areTestsRunning);
+  await ideInstance.updateState({
+    workspaceConfiguration: {
+      radonAI: {
+        areMCPTestsRunning: areTestsRunning,
+      },
+    },
+  });
+}
+
+function getIdeInstance() {
+  const ide = IDE.getInstanceIfExists();
+
+  if (!ide) {
+    throw new Error("IDE instance is not initialized. Ensure the Radon IDE panel is open.");
+  }
+
+  return ide;
+}
+
+/**
+ * Executor for `RNIDE.terminateChatToolTest` VSCode command.
+ * Terminates ongoing MCP tool tests, which were initiated by `RNIDE.testChatToolUsage` VSCode command.
+ */
+export async function terminateChatToolTest() {
+  const ideInstance = getIdeInstance();
+  await setTestStatus(false, ideInstance);
+}
+
+/**
+ * Executor for `RNIDE.testChatToolUsage` VSCode command.
+ * Temporarily takes control over the AI chat tab, testing its responses to various prompts.
+ * Running this command may interfere with other VSCode functionalities as well.
+ */
+export async function testChatToolUsage(): Promise<void> {
+  const ideInstance = getIdeInstance();
+  const runStatus: ChatTestResult[] = [];
+
+  await setTestStatus(true, ideInstance);
+
+  const fail = (testCase: ChatTestCase, cause: string) => {
+    runStatus.push({
+      cause,
+      success: false,
+      prompt: testCase.prompt,
+    });
+  };
+
+  const success = (testCase: ChatTestCase) => {
+    runStatus.push({
+      cause: null,
+      success: true,
+      prompt: testCase.prompt,
+    });
+  };
+
+  // - `showInformationMessage` cannot be programmatically dismissed
+  // - `showQuickPick` is a list-selection - does not look right
+  // - `createStatusBarItem` looks good, and can be dismissed both programmatically and by the user
+  const statusBar = window.createStatusBarItem(StatusBarAlignment.Left, 0);
+  statusBar.command = "RNIDE.terminateChatToolTest";
+  statusBar.text = "$(debug-stop) MCP tests running — Terminate";
+  statusBar.tooltip = "Click to terminate running E2E tests";
+  statusBar.color = new ThemeColor("statusBar.foreground");
+  statusBar.backgroundColor = new ThemeColor("statusBarItem.errorBackground");
+  statusBar.show();
+
+  const dir = await mkdtemp(path.join(tmpdir(), "radon-chat-exports-"));
+
+  for (const testCase of testCases) {
+    await clearEdits();
+
+    await commands.executeCommand("workbench.action.chat.newChat");
+    await commands.executeCommand("workbench.action.chat.openagent", testCase.prompt);
+
+    const shouldContinue = await awaitTestTerminationOrTimeout(ideInstance, 10_000);
+
+    if (!shouldContinue) {
+      fail(testCase, "User input: Test was terminated early.");
+      break;
+    }
+
+    const filepath = path.join(dir, randomBytes(8).toString("hex") + ".json");
+
+    await commands.executeCommand("workbench.action.chat.export", Uri.parse(filepath));
+
+    let chatData;
+    try {
+      const exportedText = readFileSync(filepath).toString();
+      chatData = JSON.parse(exportedText) as ChatData;
+    } catch {
+      fail(testCase, "Internal error: `workbench.action.chat.export` did not work.");
+      continue;
+    }
+
+    if (chatData.requests.length === 0) {
+      fail(testCase, "Internal error: `workbench.action.chat.openagent` did not work.");
+      continue;
+    }
+
+    if (chatData.requests.length > 1) {
+      fail(testCase, "Internal error: `workbench.action.chat.newChat` did not work.");
+      continue;
+    }
+
+    const responses = chatData.requests[0].response;
+
+    const toolCalls = responses.filter((response) => isToolCallResponse(response));
+
+    if (toolCalls.length === 0) {
+      fail(testCase, "No tools were called.");
+      continue;
+    }
+
+    const otherCalledTools = [];
+    let wasExpectedToolCalled = false;
+
+    for (const toolCall of toolCalls) {
+      if (testCase.allowedToolIds.includes(toolCall.toolId)) {
+        wasExpectedToolCalled = true;
+        success(testCase);
+        break;
+      }
+
+      otherCalledTools.push(toolCall.toolId);
+    }
+
+    if (!wasExpectedToolCalled) {
+      const expected = `Expected: ${testCase.allowedToolIds.join(" | ")}`;
+      const received = `Received: ${otherCalledTools.join(", ")}`;
+      const cause = `${expected}. ${received}`;
+      fail(testCase, cause);
+    }
+  }
+
+  await setTestStatus(false, ideInstance);
+
+  statusBar.hide();
+  statusBar.dispose();
+
+  rm(dir, { recursive: true }).catch((_e) => {
+    // silence the errors, it's fine
+  });
+
+  await clearEdits();
+
+  const failReasons = runStatus
+    .map((v) => `${v.success ? " OK " : "FAIL"}${v.cause !== null ? ` | Error: ${v.cause}` : ""}`)
+    .join("\n");
+
+  const correctCount = runStatus
+    .map((v) => (v.success ? 1 : 0) as number)
+    .reduce((acc, v) => v + acc);
+
+  const totalCount = runStatus.length;
+  const correctPercent = ((correctCount / totalCount) * 100).toFixed(1);
+
+  const response = `\n=== AI TEST RESULTS ===\n${failReasons}\n# TOTAL CORRECT: ${correctCount}/${totalCount} (${correctPercent}%)`;
+  Logger.log(response);
+}
diff --git a/packages/vscode-extension/src/ai/tests/chatTestCases.ts b/packages/vscode-extension/src/ai/tests/chatTestCases.ts
new file mode 100644
index 000000000..3d1451b03
--- /dev/null
+++ b/packages/vscode-extension/src/ai/tests/chatTestCases.ts
@@ -0,0 +1,94 @@
+import { ChatTestCase } from "./models";
+
+export const testCases: ChatTestCase[] = [
+  {
+    prompt: "How to use Shared Element Transitions in Reanimated 4?",
+    allowedToolIds: ["query_documentation"],
+  },
+  {
+    prompt: "How to use SETs in Reanimated?",
+    allowedToolIds: ["query_documentation"],
+  },
+  {
+    prompt: "Implement an example interaction with a local LLM in my app.",
+    allowedToolIds: ["query_documentation"],
+  },
+  {
+    prompt: "Add LLM chat to my app.",
+    allowedToolIds: ["query_documentation"],
+  },
+
+  {
+    prompt: "My button in the center of the screen is malformed.",
+    allowedToolIds: ["view_component_tree", "view_screenshot"],
+  },
+  {
+    prompt: "The orange button is ugly. Fix it.",
+    allowedToolIds: ["view_component_tree", "view_screenshot"],
+  },
+
+  {
+    prompt: "Restart the app.",
+    allowedToolIds: ["reload_application"],
+  },
+  {
+    prompt: "The app is frozen. Can you reset it?",
+    allowedToolIds: ["reload_application"],
+  },
+
+  {
+    prompt: "Why did the app just crash?",
+    allowedToolIds: ["view_application_logs"],
+  },
+  {
+    prompt: "Are there any errors in the logs?",
+    allowedToolIds: ["view_application_logs"],
+  },
+  {
+    prompt: "Debug the error thrown when I clicked the login button.",
+    allowedToolIds: ["view_application_logs", "view_component_tree"],
+  },
+
+  {
+    prompt: "Does the layout look broken to you?",
+    allowedToolIds: ["view_screenshot"],
+  },
+  {
+    prompt: "I think the text is being cut off on the right side.",
+    allowedToolIds: ["view_screenshot"],
+  },
+  {
+    prompt: "Verify if the dark mode colors are applied correctly.",
+    allowedToolIds: ["view_screenshot"],
+  },
+  {
+    prompt: "Take a look at the current screen.",
+    allowedToolIds: ["view_screenshot"],
+  },
+
+  {
+    prompt: "What is the hierarchy of the current screen?",
+    allowedToolIds: ["view_component_tree"],
+  },
+  {
+    prompt: "Show me the props passed to the Header component.",
+    allowedToolIds: ["view_component_tree"],
+  },
+  {
+    prompt: "Is the 'Submit' button currently inside a SafeAreaView?",
+    allowedToolIds: ["view_component_tree"],
+  },
+  {
+    prompt: "Find the component ID for the bottom navigation bar.",
+    allowedToolIds: ["view_component_tree"],
+  },
+
+  {
+    prompt: "Why is the banner not showing up?",
+    allowedToolIds: ["view_component_tree", "view_application_logs", "view_screenshot"],
+  },
+  {
+    prompt: "Inspect the padding on the user profile card.",
+    allowedToolIds: ["view_component_tree", "view_screenshot"],
+  },
+];
diff --git a/packages/vscode-extension/src/ai/tests/models.ts b/packages/vscode-extension/src/ai/tests/models.ts
new file mode 100644
index 000000000..7ec2100f2
--- /dev/null
+++ b/packages/vscode-extension/src/ai/tests/models.ts
@@ -0,0 +1,37 @@
+export interface ChatData {
+  requests: Request[];
+}
+
+export interface Request {
+  response: Response[];
+}
+
+export type Response = ToolCallResponse | UnknownResponse;
+
+export interface UnknownResponse {
+  // `Exclude<string, "literal">` resolves to `string` (does not work)
+  kind: unknown;
+}
+
+export type AllowedToolId =
+  | "query_documentation"
+  | "view_screenshot"
+  | "view_component_tree"
+  | "view_application_logs"
+  | "reload_application";
+
+export interface ToolCallResponse {
+  kind: "toolInvocationSerialized";
+  toolId: AllowedToolId;
+}
+
+export interface ChatTestCase {
+  prompt: string;
+  allowedToolIds: AllowedToolId[];
+}
+
+export interface ChatTestResult {
+  prompt: string;
+  success: boolean;
+  cause: string | null;
+}
diff --git a/packages/vscode-extension/src/common/State.ts b/packages/vscode-extension/src/common/State.ts
index 7e0456af7..9f82da74b 100644
--- a/packages/vscode-extension/src/common/State.ts
+++ b/packages/vscode-extension/src/common/State.ts
@@ -92,6 +92,7 @@ export type GeneralSettings = {
 
 export type RadonAISettings = {
   enableRadonAI: boolean;
+  areMCPTestsRunning: boolean;
 };
 
 export type UserInterfaceSettings = {
@@ -624,6 +625,7 @@ export const initialState: State = {
     },
     radonAI: {
       enableRadonAI: true,
+      areMCPTestsRunning: false,
     },
     userInterface: {
       panelLocation: "tab",
diff --git a/packages/vscode-extension/src/extension.ts b/packages/vscode-extension/src/extension.ts
index b1424dea2..86090b23b 100644
--- a/packages/vscode-extension/src/extension.ts
+++ b/packages/vscode-extension/src/extension.ts
@@ -40,6 +40,7 @@ import { AdminRestrictedFunctionalityError, PaywalledFunctionalityError } from "
 import { registerRadonAI } from "./ai/mcp/RadonMcpController";
 import { MaestroCodeLensProvider } from "./providers/MaestroCodeLensProvider";
 import { removeLicense } from "./utilities/license";
+import { terminateChatToolTest, testChatToolUsage } from "./ai/tests/aiChatTester";
 import { getTelemetryReporter } from "./utilities/telemetry";
 import { getEditorType } from "./utilities/editorType";
 
@@ -325,6 +326,15 @@ export async function activate(context: ExtensionContext) {
   context.subscriptions.push(
     commands.registerCommand("RNIDE.removeLicense", removeLicenseWithConfirmation)
   );
+
+  context.subscriptions.push(
+    commands.registerCommand("RNIDE.testChatToolUsage", testChatToolUsage)
+  );
+
+  context.subscriptions.push(
+    commands.registerCommand("RNIDE.terminateChatToolTest", terminateChatToolTest)
+  );
+
   // Debug adapter used by custom launch configuration, we register it in case someone tries to run the IDE configuration
   // The current workflow is that people shouldn't run it, but since it is listed under launch options it might happen
   // When it does happen, we open the IDE panel and restart the app.
diff --git a/packages/vscode-extension/src/utilities/workspaceConfiguration.ts b/packages/vscode-extension/src/utilities/workspaceConfiguration.ts
index d07fc92ee..29cfb0d86 100644
--- a/packages/vscode-extension/src/utilities/workspaceConfiguration.ts
+++ b/packages/vscode-extension/src/utilities/workspaceConfiguration.ts
@@ -19,7 +19,10 @@ const WorkspaceConfigurationKeyMap = {
     enableExperimentalElementInspector: "general.enableExperimentalElementInspector",
     inspectorExcludePattern: "general.inspectorExcludePattern",
   },
-  radonAI: { enableRadonAI: "radonAI.enabledBoolean" },
+  radonAI: {
+    enableRadonAI: "radonAI.enabledBoolean",
+    areMCPTestsRunning: "radonAI.areMCPTestsRunning",
+  },
   userInterface: {
     panelLocation: "userInterface.panelLocation",
     showDeviceFrame: "userInterface.showDeviceFrame",
@@ -58,6 +61,8 @@ export function getCurrentWorkspaceConfiguration(config: WorkspaceConfiguration)
     radonAI: {
       enableRadonAI:
         config.get<boolean>(WorkspaceConfigurationKeyMap.radonAI.enableRadonAI) ?? true,
+      areMCPTestsRunning:
+        config.get<boolean>(WorkspaceConfigurationKeyMap.radonAI.areMCPTestsRunning) ?? true,
     },
     userInterface: {
       panelLocation: