diff --git a/src/agents/__init__.py b/src/agents/__init__.py
index e3b34d244b..9474a24251 100644
--- a/src/agents/__init__.py
+++ b/src/agents/__init__.py
@@ -176,6 +176,7 @@
     resolve_computer,
     tool_namespace,
 )
+from .tool_context import ToolContext
 from .tool_guardrails import (
     ToolGuardrailFunctionOutput,
     ToolInputGuardrail,
@@ -413,6 +414,7 @@ def enable_verbose_stdout_logging():
     "CompactionItem",
     "AgentHookContext",
     "RunContextWrapper",
+    "ToolContext",
     "TContext",
     "RunErrorDetails",
     "RunErrorData",
diff --git a/src/agents/run_context.py b/src/agents/run_context.py
index df7047eb38..453bb5477e 100644
--- a/src/agents/run_context.py
+++ b/src/agents/run_context.py
@@ -61,6 +61,9 @@ class RunContextWrapper(Generic[TContext]):
     tool_input: Any | None = None
     """Structured input for the current agent tool run, when available."""
 
+    _generated_items: list[Any] = field(default_factory=list, repr=False)
+    """Internal: conversation items generated so far in the current run."""
+
     @staticmethod
     def _to_str_or_none(value: Any) -> str | None:
         if isinstance(value, str):
diff --git a/src/agents/run_internal/turn_resolution.py b/src/agents/run_internal/turn_resolution.py
index e7c059c701..f98d28ecfd 100644
--- a/src/agents/run_internal/turn_resolution.py
+++ b/src/agents/run_internal/turn_resolution.py
@@ -1890,6 +1890,10 @@ async def get_single_step_result_from_response(
 
     tool_use_tracker.record_processed_response(item_agent, processed_response)
 
+    # Expose the conversation history so tools can inspect prior items via
+    # ToolContext.conversation_history (addresses #904).
+    context_wrapper._generated_items = list(pre_step_items)
+
     if event_queue is not None and processed_response.new_items:
         handoff_items = [
             item for item in processed_response.new_items if isinstance(item, HandoffCallItem)
diff --git a/src/agents/tool_context.py b/src/agents/tool_context.py
index 7ee140e8a9..2df910ae73 100644
--- a/src/agents/tool_context.py
+++ b/src/agents/tool_context.py
@@ -72,6 +72,7 @@ def __init__(
         turn_input: list[TResponseInputItem] | None = None,
         _approvals: dict[str, _ApprovalRecord] | None = None,
         tool_input: Any | None = None,
+        _generated_items: list[Any] | None = None,
     ) -> None:
         """Preserve the v0.7 positional constructor while accepting new context fields."""
         resolved_usage = Usage() if usage is _MISSING else cast(Usage, usage)
@@ -81,6 +82,7 @@ def __init__(
             turn_input=list(turn_input or []),
             _approvals={} if _approvals is None else _approvals,
             tool_input=tool_input,
+            _generated_items=list(_generated_items or []),
         )
         self.tool_name = (
             _assert_must_pass_tool_name() if tool_name is _MISSING else cast(str, tool_name)
@@ -104,6 +106,16 @@ def __init__(
         self.agent = agent
         self.run_config = run_config
 
+    @property
+    def conversation_history(self) -> list[Any]:
+        """The items generated so far in the current agent run.
+
+        This is a snapshot of the conversation history at the time the tool was invoked,
+        allowing tools to inspect prior messages, tool calls, and other items produced
+        during the run. Returns a copy so that mutations do not affect the run state.
+        """
+        return list(self._generated_items)
+
     @property
     def qualified_tool_name(self) -> str:
         """Return the tool name qualified by namespace when available."""
diff --git a/tests/test_tool_context.py b/tests/test_tool_context.py
index a4579e8fb4..de5e1509bb 100644
--- a/tests/test_tool_context.py
+++ b/tests/test_tool_context.py
@@ -353,3 +353,66 @@ async def on_invoke_tool(
     assert captured_context is not None
     assert not isinstance(captured_context, ToolContext)
     assert captured_context.tool_input == {"city": "Tokyo"}
+
+
+def test_conversation_history_defaults_to_empty() -> None:
+    """conversation_history is empty when no generated items are set."""
+    tool_ctx: ToolContext[None] = ToolContext(
+        context=None,
+        tool_name="tool",
+        tool_call_id="call-1",
+        tool_arguments="{}",
+    )
+    assert tool_ctx.conversation_history == []
+
+
+def test_conversation_history_populated_via_wrapper() -> None:
+    """conversation_history is populated from RunContextWrapper._generated_items."""
+    wrapper: RunContextWrapper[None] = RunContextWrapper(context=None)
+    wrapper._generated_items = ["item_a", "item_b"]
+
+    tool_call = ResponseFunctionToolCall(
+        type="function_call",
+        name="my_tool",
+        call_id="call-10",
+        arguments="{}",
+    )
+    tool_ctx = ToolContext.from_agent_context(
+        wrapper,
+        tool_call_id="call-10",
+        tool_call=tool_call,
+    )
+    assert tool_ctx.conversation_history == ["item_a", "item_b"]
+
+
+def test_conversation_history_returns_copy() -> None:
+    """Mutating the returned list must not affect the underlying state."""
+    wrapper: RunContextWrapper[None] = RunContextWrapper(context=None)
+    wrapper._generated_items = ["item_a"]
+
+    tool_call = ResponseFunctionToolCall(
+        type="function_call",
+        name="my_tool",
+        call_id="call-11",
+        arguments="{}",
+    )
+    tool_ctx = ToolContext.from_agent_context(
+        wrapper,
+        tool_call_id="call-11",
+        tool_call=tool_call,
+    )
+    history = tool_ctx.conversation_history
+    history.append("extra")
+    assert tool_ctx.conversation_history == ["item_a"]
+
+
+def test_conversation_history_via_constructor() -> None:
+    """conversation_history can be set via _generated_items kwarg."""
+    tool_ctx: ToolContext[None] = ToolContext(
+        context=None,
+        tool_name="tool",
+        tool_call_id="call-12",
+        tool_arguments="{}",
+        _generated_items=["x", "y"],
+    )
+    assert tool_ctx.conversation_history == ["x", "y"]