Writing / MCP

Notes from the diagnostic boundary.

Practical writing about MCP, coding-agent feedback, caching, and the choices that make local tools trustworthy.

11 August 20266 min read

Improving MCP tool descriptions from a B to an A grade

Glama grades MCP servers on six rubric dimensions. Signalint scored B tier before v0.3.4. Here is what the rubric actually measures, what was missing, and what the rewrite changed.

MCP
Tool descriptions
TDQS

Read article

2 August 20267 min read

Why coding agents need compact diagnostics

MCP can carry a full compiler transcript. That does not mean it should. Useful diagnostic context is bounded, attributable, and shaped around the next decision.

MCP
Diagnostics
Agent tooling

Read article