contextburn 0.2.1

Listed in the official MCP Registry as ai.arsentev/contextburn, verified through the author's domain.

Changed

• MCP Registry name is ai.arsentev/contextburn.
• README and citation metadata point to the concept DOI 10.5281/zenodo.22712985, which always resolves to the latest version.

claude mcp add contextburn -- uvx contextburn mcp

contextburn 0.2.0

Adds an MCP server, so a coding agent can read its own run efficiency mid-session.

Added

• contextburn mcp starts a dependency-free MCP server (stdio) with two tools: run_efficiency returns the shares as structured data, spend_breakdown returns the full report.
• contextburn --efficiency [hours] prints run efficiency as JSON.

Changed

• The report, the JSON output and the MCP server share one efficiency calculation.

pip install contextburn
claude mcp add contextburn -- uvx contextburn mcp

No network calls. MIT licensed.

contextburn 0.1.1

First release archived on Zenodo with a DOI.

contextburn reads the transcripts Claude Code already writes on your machine and reports run efficiency: the share of paid tokens that became model output, versus the agent re-reading context it had already sent.

Added

• CHANGELOG.md and CONTRIBUTING.md.

The code is the same as 0.1.0, which was published before the Zenodo integration was enabled and therefore has no DOI of its own.

The efficiency measure comes from a controlled experiment: https://doi.org/10.5281/zenodo.22699668

contextburn 0.1.0

First public release.

contextburn reads the transcripts Claude Code already writes on your machine and reports run efficiency: the share of paid tokens that became model output, versus the agent re-reading context it had already sent.

• Two efficiency shares: by tokens (a property of how agents work) and cost-weighted (a property of how you run sessions).
• Usage records deduplicated by message id with an element-wise maximum, so streaming snapshots are neither double-counted nor truncated.
• English interface by default; Russian via CONTEXTBURN_LANG=ru or ~/.config/contextburn/lang.
• macOS menu-bar app in app/main.swift.
• Citation metadata in CITATION.cff and codemeta.json.

No network calls. MIT licensed.

The efficiency measure comes from a controlled experiment: https://doi.org/10.5281/zenodo.22699668

Formerly published as tokmon; TOKMON_* environment variables still work.