Run efficiency for coding agents: share of paid tokens that became output, not context re-reading.
ai.arsentev/contextburn — MCP Server Model Context Protocol
ai.arsentev/contextburn is an MCP server that supports “Run efficiency for coding agents.” Its description defines a specific efficiency metric: the share of paid tokens that became output, rather than tokens lost to context re-reading.
🛠️ Key Features
Provides a “run efficiency” concept for coding agents
Focuses on the ratio of paid tokens that become output
🚀 Use Cases
Evaluating coding-agent runs by token-to-output efficiency
Comparing runs where context re-reading occurs vs. productive output
⚡ Developer Benefits
Supports measurement of how efficiently paid tokens translate into output
Enables attention to context re-reading as an efficiency cost
⚠️ Limitations
Available documentation data includes only the metric definition and does not describe additional tools or behavior
contextburn reads the transcripts Claude Code already writes on your machine and tells you what
share of the tokens you paid for became model output — and how much was the agent re-reading
context it had already sent.
Token counters answer "how much did I spend?". This answers "how much of it was work?" — a
normalised share, so it can be compared across sessions, models and ways of working.
Real output over the session logs of the 36 runs behind the U-curve report — nothing else on the machine. Video with DOI: 10.5281/zenodo.22713920. The runs themselves are open: Hugging Face (DOI 10.57967/hf/10366) · Kaggle · OSF (DOI 10.17605/OSF.IO/5QTWY).
Why two numbers
By tokens the share barely moves. Every agent step resends the accumulated context, so
re-reading dominates whatever you do — it describes the agent.
Cost-weighted the share does move, because cached reads are priced far below fresh input and
output. It depends on how you run sessions — it describes you.
Reads local Claude Code transcripts (~/.claude/projects/**/*.jsonl). Nothing leaves the
machine — no network calls at all.
Deduplicates usage records by message id and keeps the element-wise maximum. A streaming runtime
writes an early snapshot and a final record for the same call: counting both double-counts it,
keeping only the first halves the output.
Weights the cost share with per-model prices kept at the top of bin/contextburn. Update them
there when they change.
Commands
command
what it shows
contextburn
what is burning tokens right now
contextburn detail [hours]
run efficiency, sessions, and what specifically inflated the context
contextburn window
the current 5-hour subscription window
contextburn --json
machine-readable state (used by the menu-bar app)
contextburn --probe <hours>
raw JSON dump of the parsed sessions
contextburn --efficiency [hours]
run efficiency as JSON
contextburn mcp
start the MCP server
Configuration
setting
default
meaning
CONTEXTBURN_LANG or ~/.config/contextburn/lang
en
interface language: en or ru
CONTEXTBURN_DAY_START
6
hour your day starts — the daily total resets here
CONTEXTBURN_WARN
30000000
tokens/hour that turns the menu-bar counter yellow
CONTEXTBURN_ALARM
90000000
tokens/hour that turns it red
The language file exists because the menu-bar app is launched from Finder, where environment
variables never reach it: echo ru > ~/.config/contextburn/lang switches both the app and the CLI.
MCP server
Let the agent read its own run efficiency mid-session. The package ships a dependency-free MCP
server (stdio) with two tools: run_efficiency returns the shares as structured data, and
spend_breakdown returns the full report.
bash
claude mcp add contextburn -- uvx contextburn mcp
Or install it as a Claude Code plugin, which registers the same server:
Raycast — source in editors/raycast, Store submission pending.
Menu-bar app (macOS)
app/main.swift is a small status-bar app. It polls contextburn --json once a minute and shows the
current burn rate with an hourly graph; click a bar to see that hour's breakdown.
bash
swiftc -O -o ContextBurn app/main.swift
Set CONTEXTBURN_BIN=/path/to/contextburn if the CLI is not in ~/bin or the usual Homebrew paths.
Limits
Claude Code transcripts only, for now.
The cost-weighted share is only as current as the price table in bin/contextburn.
Citing
Software DOI (all versions): 10.5281/zenodo.22712985. GitHub's "Cite this repository" button gives the
reference; metadata is in CITATION.cff.
This project was published as tokmon on its first day and renamed to avoid confusion with
unrelated tools of that name; TOKMON_* environment variables still work.
