cost-xray README

The project's README from github.com/tigerless-labs/cost-xray. As one markdown file: README.md.

cost-xray

See what your AI coding agent actually sends to the API — and what each part costs.

agents capture local only license MIT

Most usage tools read local logs. That shows the total cost of a call or session, but it misses the request-time context assembled before the model is invoked: system prompts, tool schemas, MCP blocks, tool results, cache reads/writes, and previous thinking blocks.

cost-xray captures the actual local API traffic for Claude Code and Codex, then attributes tokens and dollars back to the sources inside the request. It shows not just how much a turn cost, but why it cost that much.

Home — every session by agent · project · session, with token and dollar totals
cost-xray Home: every captured session grouped by agent · project · session, with per-row turns, cache-hit %, tokens, and cost
Drill a session — context-window occupancy (top) and per-source cost, down to each MCP server (bottom)
cost-xray Detail: context-window occupancy this turn (top) and cumulative per-source cost drilled down to each MCP server (bottom)

Requirements

Install

curl -fsSL https://raw.githubusercontent.com/tigerless-labs/cost-xray/master/install.sh | bash

The installer asks which agent(s) to capture — Claude Code, Codex, or both — and prompts even under curl … | bash.

Then open a new terminal and run claude / codex exactly as before — capture is automatic, no flags and no base-URL change. It's forward-only: runs started in that new shell are captured, not past history. Open the live TUI from anywhere:

cx

Capture runs as a background service (auto-start on boot, self-healing, port-adaptive) and doesn't change what your agent does, its results, or its cost — pause anytime with cx stop. See docs/install.md for systemd details, GUI agents (Cursor base-URL setup), manual (no-systemd) mode, and troubleshooting.

Usage

cx                  # open the live cost-xray TUI (from any directory)
cx status           # services' state, live ports, sessions captured
cx stop             # stop monitoring — proxies down; agents run direct (uncaptured)
cx start            # resume monitoring
cx restart          # restart the proxies (after a config change)
cx install          # (re)install — authoritative: installs the chosen agents, removes the rest
cx uninstall        # remove services + shell wrappers (keeps captured data)

cx is the whole CLI and works from any directory. Change port/upstream by editing ~/.cost-xray/env, then cx restart. (./run.sh <cmd> from the repo still works too — cx just calls it for you from anywhere.)

In the TUI, drill from the top down: agent → project → session → category → MCP server → tool → per-turn call → the real output. Every cell carries its cache split (read / write / fresh / output $).

Supported Agents

Agent Status Capture
Claude Code Supported reverse proxy (base-URL override) — no certificate
Codex Supported forward proxy + scoped local CA (self-healing wrapper)

The wire is decoded by a thin per-agent adapter — the only place code forks by agent (docs/architecture.md; per-agent capture + tokenizer notes under docs/providers/). Adding an agent is one small module; anything speaking the Anthropic or OpenAI-Responses wire shape is close to drop-in.

Features

Cost attribution, below the tool

cost-xray traces every token's cost — split into fresh / cache-read / cache-write / output — to the source that caused it, and down to the individual call: the cost of this Read invocation and its output, not a session sum. Everyone else aggregates — a request- or session-level total, at most grouped by tool type ("Read cost $X this session"). As far as we've found, nothing else prices below the tool, call by call.

Window occupancy

What is taking space in the context window right now — system prompt, every tool schema, MCP servers, messages, and generated output — decomposed into source-level rows. Prompt caching makes a stable 40k-token schema block cheap on cache read, but it still crowds out the code and conversation that matter; cost-xray shows you the occupancy, not just the bill.

Unused MCP waste

Configured servers and tools that are injected into every request's prefix but never actually called. They pay their tool-schema overhead on every turn — cost-xray flags the dead weight.

Tokenization accuracy

Claude's tokenizer is private — no official or open-source tokenizer exists — and calling Anthropic's count_tokens API for everything would add load and hit its limits. So for Claude, cost-xray uses an estimator plus proportional calibration: tiktoken sizes each part, the parts it mis-sizes most (thinking, tool schemas) get targeted corrections — pinned with count_tokens when you're logged in, a fixed ratio otherwise — and the rest is scaled so the calibrated total matches the provider's own usage. The total, and therefore the bill, is exact; only the split between sources in the same request is approximate. We benchmark those residuals continuously (CONTRIBUTING.md) and they're small enough for attribution work.

Coming soon: an opt-in exact mode — every part sized by full count_tokens differencing — manually enabled, for users who need maximum per-source precision.

Self-healing capture

The wrappers are per-command and self-healing: if the proxy is down, the wrapper restarts it and routes through; if it can't, the agent runs direct — never broken. Stop monitoring anytime with cx stop (agents then run direct).

Capture everything, store almost nothing

cost-xray keeps the complete raw API traffic — but a long session re-sends its whole history every turn, so the capture is hugely repetitive. We deduplicate it: each unique block (message, schema, tool result) is stored once, with a tiny per-turn delta. Full per-turn bytes rebuild on demand. Disk stays small even across million-token sessions.

Why not just read the logs?

Log-based tools (ccusage, codeburn, and similar) are excellent at local-first session analytics: they read local transcripts, classify turns by tool usage, and price the session by model, day, or task. That answers how much did I spend. cost-xray answers the lower-level question: what bytes did the model actually receive, and which source owns those tokens?

The difference is the data source. Log readers see the transcript after the agent has run — but the system prompt, injected tool schemas, MCP schemas, reminders, and provider-added blocks are assembled at request time and never written to the transcript. In real coding-agent requests, that invisible prefix can be roughly half the context or more. cost-xray reads the raw API request, so it can compute source-level tokens and attribute cost to schemas, MCP servers, tools, and message buckets.

Question Log / usage tools cost-xray wire capture
How much did the session cost? Yes Yes
Which task/tool was active? Yes Yes
System prompt and injected schemas visible? No Yes
How many tokens does each tool schema occupy? No, schemas aren't in logs Yes
Which MCP server is dead weight in the prefix? Estimate Exact
Cache read/write/fresh dollars per source/tool? No, usage is request-level Yes, by span and cache boundary
Live view of the current request window? No Yes

Reading the dashboard

cost-xray surfaces the data; you read the story. A few patterns worth knowing:

Signal you see What it might mean
A 40k-token MCP schema block on every turn A configured server crowding the prefix — drill it to see if any tool was called
Cache-read dollars dwarf fresh input Stable prefix is working; the spend is in what's new each turn
Repeated cache-write on the same source The prefix is being rewritten — something upstream of it changed
Thinking tokens dominate output cost Long reasoning turns; check whether they earned their keep
A tool's schema costs more than the tool is ever used Candidate to drop from the agent's tool-set
Big tool_result rows on Read/Bash Uncapped output bloating the window — cap it at the source

These are starting points, not verdicts. One experimental session looking odd is fine; the same pattern across weeks of work is a config issue.

How it works

cost-xray runs mitmproxy as a local capture hop and keeps all analysis off the request path.

agent ──HTTP──▶ mitmproxy ──HTTPS──▶ model API
                     │
                     └── redacted raw request/response → ~/.cost-xray/sessions/
                                      │
                                      └── materializer → TUI / cost attribution

See docs/install.md for setup and docs/architecture.md for the full pipeline.

Credits

Capture built on mitmproxy; the SSE/redaction approach was adapted from llm-interceptor. Pricing data from LiteLLM. README structure inspired by codeburn.

Built by Tigerless Labs.

License

MIT

© Tigerless · tigerless.ai · tigerless.com

cost-xray (tigerless-labs/cost-xray)

llm-observability
Searching Database…

5.6K ·GitHub
Cost-xray
What is this?cost-xray is a local tool that captures the raw API traffic between Claude Code or Codex and their providers, then breaks down exactly what's in each request and what each part costs — system prompts, tool schemas, MCP blocks, cached vs fresh tokens, and more.
Is this a logging dashboard like ccusage, or something different?It's different from log-based tools: instead of reading local transcripts after a session ends, it runs a local proxy that intercepts the actual API requests, so it sees hidden pieces like system prompts and injected tool schemas that never show up in a transcript.
How do I get it running?Run the installer (curl ... | bash, or ./install.sh if you've already cloned the repo); it asks which agent(s) to capture — Claude Code, Codex, or both. After that, just open a new terminal and run claude or codex as usual; capture starts automatically as a background service.
Do I need an API key or to change my agent's config?No — no API keys, no account, and no config changes to Claude Code or Codex are needed; capture is a transparent local proxy hop. You just need a supported agent already installed, on macOS or Linux.
Does it pick up sessions I already had running?No, it's forward-only — only sessions started in a new terminal after install get captured, not past history. Also note that for Claude, the private tokenizer means per-source token splits are estimated (calibrated against real totals), not exact.
Does it run on Windows?Not stated for Windows — only macOS and Linux are listed as supported, with GUI agents like Cursor needing manual base-URL setup.
What happens to my API keys and secrets as they pass through the capture proxy?They're redacted before anything touches disk — Authorization headers, API keys, cookies and secret-looking body fields are stripped. The proxy only binds to 127.0.0.1, sends no telemetry, and everything lives locally under ~/.cost-xray/ until you delete it.
What does the window occupancy view show?It breaks down what's currently filling the model's context window — system prompt, each tool schema, MCP servers, messages, and output — so you can see what's crowding things out even when it's cheap thanks to caching.
Can it tell me if an MCP server is wasted overhead?Yes — it flags configured MCP servers or tools whose schemas get injected into every request's prefix but are never actually called, exposing the dead weight you're paying for on every turn.
What happens if the capture proxy crashes?The per-agent wrapper is self-healing: if the proxy goes down it restarts and routes through, and if it still can't connect, the agent simply runs direct — capture never breaks your agent.
Why not just use a log-based tool like ccusage?Log readers only see the transcript after the fact, missing the system prompt, injected tool schemas, and MCP blocks assembled at request time — which can be roughly half the context or more; cost-xray reads the raw wire traffic instead.
What are similar projects?ThinkWatch-Lite, casbin-gateway, pxpipe, tokscale, codeburn, lean-ctx, headroom, billion-context
Which topics is it in?prompt-caching, llm-observability, cost-optimization, cost-tracking, ai-coding-agent, token-usage, codex-cli, reverse-proxy, tui, observability, proxy, anthropic
Creator
tigerless-labs
Size
0.4 MB
Stars
5.6K
Updated
CommitOct 15
Age
4mo
Commits · 90 Days