Claude Observability Plugin

Langfuse Observability Plugin for Claude Code

Claude Code plugin that sends Claude Code
session telemetry to Langfuse. It traces user prompts,
agent turns, model generations with their token usage and cost, thinking blocks,
tool calls, skills, subagents, and pasted or returned images.

Langfuse also documents this integration on the
Claude Code integration page.

Quick Start

Add the marketplace and install the plugin:

Restart Claude Code after installing, so that it loads the hook.

Tracing covers the claude CLI and the desktop app in Code mode. Claude
Desktop Chat mode runs no Claude Code hooks and is not traced.

Prerequisites

The plugin needs uv on PATH. The hook runs as a
uv script and installs the Langfuse SDK from its own inline script metadata.
Machines without uv fall back to Python 3.10+ as python3 with
langfuse>=4.7,<5 installed.

Without a usable runtime the hook writes the reason to its log and exits. It
never blocks or slows Claude Code.

Langfuse Credentials

Configure the plugin from inside a Claude Code session. This is a Claude Code
slash command, not a shell command:

You can also pass the values during install:

Only LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY are required. Without
LANGFUSE_BASE_URL the plugin uses https://cloud.langfuse.com (EU region).
Get the keys from the API Keys page in your Langfuse project settings. The
secret key is held in your OS keychain, not in a file.

/plugin configure also lists the optional settings, such as the user ID,
debug logging, and image capture, with their defaults.

Updating

Automatic updates are off for marketplaces outside Anthropic's, so you run the
update yourself.

Restart Claude Code afterwards. claude plugin list shows the installed
version.

Troubleshooting

The hook explains nearly every…