Skip to main content
AI/MLjavimosch

context-mode

Use context-mode (via supercli MCP) for large codebase analysis (50+ files), long command outputs, or repeated searches. Reduces token usage by up to 98%. NOT for simple commands or small file sets.

Stars
37
Source
javimosch/supercli
Updated
2026-05-31
Slug
javimosch--supercli--context-mode
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/javimosch/supercli/HEAD/.agents/skills/context-mode/SKILL.md -o .claude/skills/context-mode.md

Drops the SKILL.md into .claude/skills/context-mode.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

context-mode Decision Guide

context-mode is an MCP server accessed via supercli. It maintains a persistent knowledge base across sessions.

Quick Decision Tree

Task size?

  • Small (1-10 files, < 500 lines output) → Use direct bash/read_file
  • Large (50+ files, 1000+ lines output) → Use context-mode via supercli

Will you search the same content multiple times?

  • Yes → Use context-mode (index once, search many times)
  • No → Use direct tools

Most Common Workflow (via supercli)

# 1. Verify context-mode is available
sc mcp list --json

# 2. Verify installation (daemon auto-starts)
sc mcp call --mcp-server context-mode --tool ctx_doctor --input-json '{}' --json

# 3. Explore codebase
sc mcp call --mcp-server context-mode --tool ctx_batch_execute --input-json '{
  "commands": [
    {"label": "Structure", "command": "find . -type f -name \"*.js\" | head -100"},
    {"label": "Package", "command": "cat package.json"}
  ],
  "queries": ["entry point", "dependencies", "routes"]
}' --json

# 4. Search indexed content (if needed)
sc mcp call --mcp-server context-mode --tool ctx_search --input-json '{
  "queries": ["search terms"]
}' --json

Tool Selection Guide

  • ctx_batch_execute: Default for exploration - runs commands, indexes, searches in one call
  • ctx_index + ctx_search: When you'll search the same content repeatedly
  • ctx_execute: Single long-running command (npm test, build)
  • ctx_fetch_and_index: External documentation or API specs
  • ctx_purge: Clear knowledge base between unrelated tasks
  • ctx_stats: Check token savings

When NOT to Use

  • ls -la, cat package.json → Direct bash
  • Checking 1-10 files → read_file
  • Quick duplicate check → bash + uniq -c
  • Simple grep → Direct grep or Grep tool
  • Output < 500 lines → Direct tools

Installation (if not available)

npm install -g context-mode

Then add to your harness's MCP server configuration with command: context-mode