Full trace history
Each session is stored as a trace with model turns, tool calls, spans, and a full conversation history, rather than only a local log file.
LangWatch
ClaimLangWatch tracks Claude Code sessions as traces so developers can review model turns, tool calls, token usage by class, and theoretical spend. It also supports Codex, OpenCode, Copilot, Cursor, and Pi through the same OTLP-based tracing approach.
Overview
LangWatch’s Claude Code usage page is an observability view for coding-agent sessions. It turns Claude Code runs into traces so you can inspect model turns, tool calls, token usage by class, duration, and cost in one place.
The page emphasizes two related needs: understanding how an agent behaved during a session, and understanding what that session would have cost. It also notes that the same tracing approach applies to Codex, OpenCode, Copilot, Cursor, and Pi through a shared OTLP endpoint.
What LangWatch shows for Claude Code sessions
Token accounting by class
Token usage is split into classes such as input, output, cache read, and cache write, which makes it easier to see where usage is accumulating.
Theoretical cost tracking
For bundled plans, LangWatch shows a theoretical spend so you can understand usage even when there is no per-token invoice.
Trace review and filtering
The trace interface exposes session summaries, duration, span counts, cost, model information, and filters for reviewing trace groups.
Tool-call visibility
The product can capture tool and skill activity, including Bash commands, file edits, MCP calls, and other execution spans.
MCP access to trace history
LangWatch can also publish traces back to the agent through its MCP server, allowing the agent to inspect prior runs and adapt its own workflow.
Practical ways teams use LangWatch for Claude Code
Review an individual Claude Code session
A solo developer can review a session after the fact, inspect model turns and tool calls, and see where tokens or cache reads accumulated during a long coding task.
Analyze usage patterns across a team
An engineering team can compare traces across sessions to understand how different prompts, models, or workflows affect duration and spending over time.
Centralize agent observability
A team using multiple coding agents can collect Claude Code, Codex, OpenCode, Copilot, Cursor, and Pi traces into one store for consistent review.
Feed trace history back into the agent
A workflow that depends on MCP can use LangWatch’s trace exposure to let the agent inspect previous runs and reason about its own process.
Estimate theoretical spend on bundled plans
A team on bundled subscription plans can estimate what sessions would have cost on usage-based billing before deciding how to manage usage.
Pros and Cons
Pros
- Captures full trace history instead of relying only on local logs.
- Breaks token usage into multiple classes, including cache reads and writes.
- Shows theoretical spend for bundled subscriptions, which is useful for usage analysis.
- Includes detailed execution spans for tool calls such as Bash and MCP activity.
- Supports more than one coding agent, not just Claude Code.
- Uses OpenTelemetry, which suggests a standard instrumentation path across machines.
Cons
- The source does not document every setup detail beyond the `npx langwatch claude` command and OpenTelemetry export.
- Feature coverage is partial in the source, so some dashboard, alerting, and comparison capabilities are not clearly shown on this page.
- Bundled-plan sessions are shown with theoretical cost rather than an actual per-token bill, so the cost view is informational rather than an invoice.
FAQ
How do I track Claude Code usage with LangWatch?
Run `npx langwatch claude` to configure Claude Code’s native OpenTelemetry export to LangWatch. After that, each session appears as a trace with model turns, tool calls, tokens by class, and theoretical cost.
Does this work with other coding agents besides Claude Code?
Yes. The page says Codex, OpenCode, Copilot, Cursor, and Pi use the same OTLP endpoint, so their activity can be collected in the same store with the same trace, token, and cost treatment.
What does the bundled cost label mean?
Sessions on Claude subscription plans such as Pro, Max, and Team are labeled as bundled, meaning they are not billed per token. LangWatch still shows the theoretical cost so you can see what the session would have cost on usage-based billing.
What output does LangWatch show for each session?
The trace view shows a conversation timeline with session-level summary data, spans, model usage, token totals split by class, and cost figures. The page also shows filters and trace summaries such as simplified conversations and errors.
Is LangWatch free to start?
The source describes LangWatch Cloud pricing with a free Developer plan, a Growth plan at €29 per core seat per month, and an Enterprise plan with custom pricing. It also offers self-managed deployment options.
Quick Facts
- Category
- Developer Tool / Observability
- Primary focus
- Claude Code usage tracking and trace history
- Related tools
- Codex, OpenCode, Copilot, Cursor, Pi
- Setup signal
- `npx langwatch claude`
- Pricing shape
- Free Developer plan; paid Cloud plans from €29 per core seat per month; self-managed options available
- Source domain
- langwatch.ai
LangWatch Alternatives
Orca
Orca is an Agent Development Environment for shipping with coding agents, running multiple CLI agents in parallel across isolated worktrees, with desktop and mobile workflows.
Firebase Studio
Firebase Studio is a web-based workspace for full-stack app development with Gemini-assisted coding, app previews, cloud emulators, collaboration, and browser deployment.
JetBrains
JetBrains makes developer tools for individuals and teams, including IntelliJ IDEA, PyCharm, DataGrip, WebStorm, Rider, and CLion, with built-in run/debug/test workflows.
小艺
小艺 is Huawei’s AI smart assistant for Q&A, writing, document reading, code help, image recognition, and file drag-and-drop.
EZsite AI
EZsite AI is an AI website builder that turns a URL into a fullstack React or Vue.js app with hosting, custom domains, code export, and backend features.
AI Magicx
AI Magicx is a unified AI workspace for chat, image, video, voice, music, email and developer tasks, helping teams and creators manage multiple models in one place.