cyberchitta/llm-context.py

uv tool install "llm-context>=0.6.0"
lc-init (creates .llm-context/, copies the lc-curate-context skill to .claude/skills/)

Usage

• •Human workflow: lc-init once, then lc-select and lc-context to copy formatted context to clipboard and paste into an LLM chat
• •MCP integration: add to Claude Desktop config — command 'uvx', args ['--from','llm-context','lc-mcp'], then restart Claude Desktop
• •Agent workflow (CLI): lc-outlines to explore, lc-preview <rule> to validate a rule, lc-context <rule> to generate focused context for a sub-agent
• •Agent workflow (MCP): lc_outlines(root_path, rule_name), lc_preview(root_path, rule_name), lc_missing(root_path, param_type, data, timestamp)

Key features

• ✓Composable YAML+Markdown rules in five categories: prompt (prm-), filter (flt-), instruction (ins-), style (sty-), excerpt (exc-)
• ✓Code excerpting that extracts structure while reducing tokens (15+ languages)
• ✓MCP tools lc_outlines / lc_preview / lc_missing for AI agents in chat
• ✓CLI commands: lc-init, lc-select, lc-context (-p, -m, -nt flags), lc-preview, lc-set-rule, lc-outlines, lc-missing
• ✓AI-assisted rule creation via a Claude Code skill or built-in instruction rules; deployment patterns for AI Studio, Grok, Claude Projects, and more
• ✓Project-relative path format enabling multi-project context composition

Best for

Developers and AI agents who want to share focused, task-specific project context with LLMs while keeping token usage low.

Caveats

• ⚠Not a standalone inference server — it curates and copies context; the LLM does the reasoning
• ⚠For MCP use, the AI can only access additional files the server exposes (lc_missing), not the whole repo at once