LangSmith Tracing for Claude Code plugin
The LangSmith plugin sends Claude Code conversations to LangSmith so you can see model calls, tools and subagents
- 73 ★ on the repository
- Last commit
Broad access
Why this level
- Once enabled it sends messages, tool calls and their content to the cloud LangSmith, and secret stripping is no guarantee
- Reads git metadata (repository, branch, commit, author) and keeps an API key or sign-in session in your profile
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/langsmith-tracing-for-claude-codeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Once enabled it sends messages, tool calls and their content to the cloud LangSmith, and secret stripping is no guarantee
- Reads git metadata (repository, branch, commit, author) and keeps an API key or sign-in session in your profile
This is third-party code. Review the repository files before installing.
What it does
The plugin records Claude Code work in LangSmith as a trace. Each model call lands there with the conversation so far, the reply, the model name, the provider and token counts. Tool calls are saved with input and output, skill calls with the name of the skill that ran, and subagents appear nested under the turn that started them. When a tool touches a file in a git repository, the record is tagged with the repository name, branch, commit and author. A subagent uploads after it finishes, so a cancelled one drops its steps. Tracing is off by default. `/langsmith-tracing:setup [project] [url]` or the file `~/.claude/langsmith.json` with `enabled` and `project` or `agent_id` turns it on. Sign-in to LangSmith goes through `langsmith auth login` or an `api_key`; with several options the plugin's own key wins, then the CLI sign-in, then a generic `LANGSMITH_API_KEY`. `/langsmith-tracing:trace` returns a link to the current conversation's trace. For privacy there are `/langsmith-tracing:mute` and `unmute`: a muted turn keeps its structure and loses the text. The `defaultMuted` setting mutes every conversation at once. Secrets are stripped before upload unless `redact` is off: API keys, JWTs, PEM blocks and common credential patterns. Admins set policy with a mandatory `~/.langsmith-plugins/managed.json` file, and `strict_auth` refuses prompts until sign-in works.
Who it is for. Teams building on LangChain and LangSmith who want to see in traces what Claude Code actually did, and admins rolling the plugin out across an organization.
Good fit when
- The agent made an unexpected edit and you need to trace which model and tool calls led to it
- The team wants traces of all Claude Code sessions in one LangSmith project for analysis
- An admin needs tracing enabled for everyone with the project and workspace pinned through MDM
- Some conversations hold sensitive data and must be muted with `/langsmith-tracing:mute`
Not a fit when
- Data must not leave the machine: enabled tracing uploads messages and tool inputs and outputs
- You have no LangSmith account: the plugin needs a sign-in or a key
- You need tracing inside Claude Code itself with no external service: the plugin writes to LangSmith
Example request
Install the LangSmith plugin, turn tracing on for the claude-code project and give me a link to this conversation's traceLimitations
- Windows and Linux need Node.js 20 or newer. On Mac the plugin carries its own build, with Node as a fallback.
- With tracing on, messages, tool inputs and outputs and metadata go to LangSmith. Secret stripping is no guarantee, so leave tracing off if in doubt.
- Muting applies from the next turn and does not delete earlier uploads.
- A cloned repository can switch tracing on through `.claude/langsmith.json`, so check the settings of repositories you did not write.
How to disable. Turn tracing off with enabled: false in `~/.claude/langsmith.json` or `TRACE_TO_LANGSMITH=false`. Remove the plugin entirely through `/plugin` in Claude Code or with claude plugin uninstall langsmith-tracing@langsmith-claude-code-plugins. Uploaded traces stay in LangSmith and are deleted there.
FAQ
How do I install the LangSmith plugin for Claude Code?
In Claude Code run `/plugin marketplace add langchain-ai/langsmith-claude-code-plugins`, then `/plugin install langsmith-tracing@langsmith-claude-code-plugins` and `/reload-plugins`. Auto-update is off for this marketplace by default.
How do I turn tracing on?
Run `/langsmith-tracing:setup` with a project name or LangSmith URL, or set `{ "enabled": true, "project": "shared-claude" }` in `~/.claude/langsmith.json`. With no setting traces go to the `claude-code` project.
What reaches LangSmith?
Messages, tool inputs and outputs, metadata, token usage and subagent structure. Secrets are stripped before upload unless `redact` is off. Muted turns upload only structure.
How do I hide a conversation with sensitive data?
Run `/langsmith-tracing:mute` and wait for the confirmation before sending sensitive text. `/langsmith-tracing:unmute` brings upload back. The `defaultMuted` setting mutes all conversations by default.
How can an organization turn tracing on for everyone?
Install the plugin through Claude Code managed settings and place `~/.langsmith-plugins/managed.json` on each machine, usually with an MDM tool such as Jamf, Kandji or Intune. The file must be root-owned and not writable by others.
Discussion
No questions or reviews yet.
Related
PluginsA terminal dashboard for Claude Code and Codex: a tree of agents with live status, a tool-call feed and context headroom.
OpenTelemetry-based CLI tracing for Claude Code, Cursor and Codex sessions, showing what the agent actually did and what it cost
Claude Watch draws a localhost map of project logic, shows Claude Code tool calls live and answers questions
Claude Code plugins by shokai: Codex second opinions, pull request review, repository security patrol and a factory mode.
A transparent wrapper for Claude Code and Codex: a local gateway records run events and trajectories without changing the agent.
Local proxy between Claude Code and Anthropic that fixes prompt-cache misses, stabilizes the request prefix and shows quota use in the status line
README in short
The project page presents the plugin as a way to send Claude Code conversations to LangSmith to see what the agent did. It lists requirements, install for yourself and for an organization, a managed settings file with permission checks and three sign-in ways. Enabling tracing, trace contents, hiding conversations, common problems, a privacy section and a settings table with environment variables follow.