context-mode hits 30,000 stars with a local sandbox for coding agents' context windows

context-mode, an MCP server by mksglu for trimming AI coding agents' context-window consumption, has passed 30,000 GitHub stars. The approach: instead of dumping raw tool output into the context window, it sandboxes the output and hands the agent a compact summary, with the author reporting up to 98% less context consumed. It also keeps a persistent local memory of edits, git operations, tasks, errors, and decisions in SQLite with full-text search, so sessions survive compaction and resume without relearning the codebase. Installation on Claude Code is a plugin command, with hooks or MCP-only setups for sixteen other platforms. Everything runs locally with no telemetry, though note the license is the Elastic License v2, which is source-available rather than fully open source.

Summarized by the Neuricon wire. Every claim links back to its primary source where available.