Inside the read-only semantic-analysis plugin for local models
How Agent Harness's slim, read-only MCP plugin gives local models semantic code analysis without granting them write access to anything.
By Agent Software
Scheduled article
This post is scheduled to publish in full on December 28, 2026. The final URL is already live so social previews, search crawlers, and future readers all resolve to the same article.
A local model doesn't need write access to be useful for analysis — Harness's semantic-analysis plugin is deliberately read-only, and deliberately small.
- Final publish date: December 28, 2026
- Primary categories: Testing
- Tags: #suite #agent-harness #local-models
In the meantime, the related posts below cover the adjacent product and engineering context this article builds on.
