Catches drift before the PR
When the model starts padding reasoning with hedges, you know — before review time costs you hours.
mycelitree
An MCP server that measures the space between your CLI and your tools.
Catches drift. Flags performed confidence. Surfaces when reviews are less independent than they look.
01 — what it catches
Three things every other MCP misses.
Measures effective review coverage
If three reviewers use the same tool, the coverage isn't 3×. We tell you what it is.
Flags performed confidence
Rising certainty without new evidence is visible to the measurement. Not to most readers.
02 — before / after
The same text, measured.
as you read it
Looking at audit.js, the file is 245 lines and runs entirely client-side. It's worth noting that several distinct metrics contribute to the composite. The graph contains 80,019 entities and 106,311 relationships as of 2026-04-26. The methodology is documented at https://mycelitree.com/api/audit and is open for inspection. Generally speaking, no single metric captures everything, but the composite has held up across the test corpus.
as mycelitree sees it
Looking at audit.js, the file is 245 lines and runs entirely client-side. It's worth noting that several distinct metrics contribute to the composite. The graph contains 80,019 entities and 106,311 relationships as of 2026-04-26. The methodology is documented at https://mycelitree.com/api/audit and is open for inspection. Generally speaking, no single metric captures everything, but the composite has held up across the test corpus.
performed confidence (hedging)
certainty without new evidence
03 — install
Run it on your own outputs.
$ npx mycelitree-mcp init
$ claude mcp add mycelitree
$ claude mcp add mycelitree
install · coming soon
Five tools. One server.
check_drift · measure_coverage · flag_confidence · audit_independence · trace_provenance
Install the MCP to see it on your own outputs. All computation runs client-side. Read the docs →