Side-by-side comparison of two Model Context Protocol servers — install paths, tools, quality signals, and directory engagement so you can pick the right one for Claude, Cursor, and other MCP clients.
Local-first CLI and MCP wrapper for AI coding agents. Routes shell commands through sage run --, stores command history locally, and returns compressed terminal output to reduce context noise.
Answers "what did that command print?" from your real terminal history: a per-command PTY session recorder (OSC 133) stores every command's full output, exit code, cwd, and duration in local SQLite/FTS5, and the MCP server searches, retrieves, and diffs those outputs — secret redaction on by default before anything reaches the model.
Quality signal
52/100 (Fair)
44/100 (Fair)
Install path
npx · high
npx · low
Engagement
2 0 0 10
1 0 0 6
Tools
Runs commands and returns compressed terminal output summariesKeeps full raw output and command history locallySupports shortcuts for common commands (e.g., pytest, npm test, git status)Tracks token savings and command success metricsOptional cloud mode for sending aggregate proof metricsCompatible with multiple AI coding agents and CI scripts
Per-command PTY session recording with OSC 133 semantic-prompt marksLocal SQLite database with FTS5 trigram full-text searchZstd-compressed command outputs with configurable head/tail captureAutomatic redaction of secrets before output reaches the modelCross-platform shell integration for Bash, Zsh, Fish, PowerShellNo cloud or telemetry; all data stored locally in a single SQLite file
