langchain - 💡(How to fix) Fix Add tool_call_id to on_tool_start event's data

Submission checklist

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

The on_tool_start stream event currently exposes name, tags, run_id, metadata, parent_ids, and data.input — but not tool_call_id.

LangChain already tracks tool_call_id for each tool run: BaseTool.arun passes it to the callback manager, _AstreamEventsCallbackHandler stores it in run_info during on_tool_start (since #33731), and it's surfaced on on_tool_error.data (also #33731). The data is right there; it's just not emitted on on_tool_start.

import asyncio
from langchain.agents import create_agent
from langchain_core.tools import tool

@tool
def slow_lookup(query: str) -> str:
    """A tool that takes a bit of time so we can observe its start event."""
    return f"result for {query}"

agent = create_agent(model=..., tools=[slow_lookup])

async def demonstrate_issue():
    async for event in agent.astream_events(
        {"messages": "look up alpha and beta in parallel"},
        version="v2",
    ):
        if event["event"] == "on_tool_start":
            # event["data"] has "input" but no "tool_call_id"
            assert "tool_call_id" not in event["data"]

asyncio.run(demonstrate_issue())

### Use Case

Streaming UI consumers building per-tool-call panels need to attribute every on_tool_* event to the originating tool call as soon as the tool starts — not wait for on_tool_end / on_tool_error. This matters especially under parallel tool calls of the same tool (e.g., a supervisor agent dispatching several sub-agents of the same kind in one assistant turn), where name alone isn't sufficient to disambiguate which streamed events belong to which call, and falling back to run_id/parent traversal is fragile.

on_tool_error already exposes tool_call_id in its event data (added in #33731). Doing the same for on_tool_start is the natural parity.

### Proposed Solution

Mirror the on_tool_error change from #33731 on the on_tool_start send:

```python
# libs/core/langchain_core/tracers/event_stream.py — in on_tool_start
self._send(
    {
        "event": "on_tool_start",
        "data": {
            "input": inputs or {},
            "tool_call_id": kwargs.get("tool_call_id"),   # ← new
        },
        "name": name_,
        "tags": tags or [],
        "run_id": str(run_id),
        "metadata": metadata or {},
        "parent_ids": self._get_parent_ids(run_id),
    },
    "tool",
)

The kwarg is already passed in (it's used at line 311 to populate run_info); this just surfaces it on the emitted event too.

Alternatives Considered

• Reading the streamed metadata field. Possible if the BE manually splices tool_call_id into the tool's runtime config metadata via a BaseTool subclass that overrides ainvoke. Works, but every downstream agent author has to invent the same workaround. The data should be first-class on the event.
• Walking parent_ids / run_id to recover the originating tool call. Fragile — LangGraph collapses parent relationships, and run_id is separate from tool_call_id.

Additional Context

Parallel to #33597 / #33731. Same justification (stateless agents, streaming-UI tool-call mapping), same one-line surface fix.