From f2270cdb63facefb5e03aaeedb00350cedd4c402 Mon Sep 17 00:00:00 2001
From: Armin Ronacher <armin.ronacher@active-4.com>
Date: Sun, 27 Sep 2026 18:00:25 +0200
Subject: [PATCH] feat(agent): let tools return error results without throwing

AgentToolResult.isError marks a result as failed while keeping details and structuredContent, so partial output and structured failures (such as MCP isError results) reach the UI and nested callers.
---
 packages/agent/src/agent-loop.ts | 2 +-
 packages/agent/src/types.ts      | 9 ++++++++-
 2 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/packages/agent/src/agent-loop.ts b/packages/agent/src/agent-loop.ts
index 5e25c35ce..029aae278 100644
--- a/packages/agent/src/agent-loop.ts
+++ b/packages/agent/src/agent-loop.ts
@@ -936,7 +936,7 @@ async function executePreparedToolCall(
 		);
 		acceptingUpdates = false;
 		await Promise.all(updateEvents);
-		return { result, isError: false };
+		return { result, isError: result.isError === true };
 	} catch (error) {
 		acceptingUpdates = false;
 		await Promise.all(updateEvents);
diff --git a/packages/agent/src/types.ts b/packages/agent/src/types.ts
index 44617bb66..89c0ad5ab 100644
--- a/packages/agent/src/types.ts
+++ b/packages/agent/src/types.ts
@@ -447,6 +447,12 @@ export interface AgentToolResult<T = JsonValue | undefined> {
 	structuredContent?: JsonValue;
 	/** Usage from the final tool execution itself, if available. Not used for main LLM context accounting. */
 	usage?: Usage;
+	/**
+	 * Report a failure without throwing. The model sees `content` as an error result, like a thrown
+	 * error, but `details` and `structuredContent` are kept, so the UI and programmatic callers can
+	 * still read partial output or a structured failure (for example an MCP result with `isError`).
+	 */
+	isError?: boolean;
 	/**
 	 * Hint that the agent should stop after the current tool batch.
 	 * Early termination only happens when every finalized tool result in the batch sets this to true.
@@ -511,7 +517,8 @@ export interface AgentTool<TParameters extends TSchema = TSchema, TDetails = any
 	 */
 	outputSchema?: TSchema;
 	/**
-	 * Execute the tool call. Throw on failure instead of encoding errors in `content`.
+	 * Execute the tool call. Throw on failure, or return a result with `isError: true`; do not only
+	 * describe the failure in `content`.
 	 * `context` is provided by the agent loop and absent when the tool is called directly.
 	 */
 	execute: (