LangChain Webhooks and Event Dispatch (Python)

LangChain Webhooks and Event Dispatch (Python)

Overview

A team wires per-tool webhook dispatch from their LangChain agent via FastAPI BackgroundTasks — analytics is always N seconds late because BackgroundTasks fire after the HTTP response closes, not during the stream (P60). Worse: the BaseCallbackHandler they attached via .with_config(callbacks=[h]) fires on the outer agent but is dark on the subagent's tool calls — custom callbacks are not inherited by LangGraph subgraphs (P28), they must be passed via config["callbacks"] at invoke time.

Pain-catalog anchors handled here:

• P28 — Callbacks via with_config don't propagate to subgraphs
• P46 — SSE streams dropped by buffering proxies (see langchain-langgraph-streaming)
• P47 — astream_events(v2) emits thousands of events; never forward raw
• P48 — Sync invoke() inside async endpoint blocks the event loop
• P60 — BackgroundTasks fire post-response; wrong for per-event dispatch

This skill walks through an async AsyncCallbackHandler with fire-and-forget dispatch, per-target sinks for HTTP / Kafka / Redis Streams / SNS, HMAC-signed delivery with 1s/5s/30s retry and DLQ, idempotency keys = `run_id + event_type

• step_index, andconfig["callbacks"]wiring that makes subagent calls visible. Typical webhook latency budget: <500ms per event. Pin:langchain-core 1.0.x, langgraph 1.0.x. Scope: server-to-server dispatch only — UI streaming is in langchain-langgraph-streaming`.

Prerequisites

• Python 3.10+
• langchain-core >= 1.0, < 2.0, langgraph >= 1.0, < 2.0
• httpx >= 0.27 for async HTTP (or aiohttp)
• One of: aiokafka, redis[hiredis] >= 5, aioboto3 (per target)
• An event sink — a webhook endpoint, Kafka topic, Redis Stream, or SNS topic
• A shared secret (for HMAC) stored in your secret manager, not env

Instructions