Massive Context MCP

#!/usr/bin/env bash # Hook script to suggest RLM tools for large file reads # This is a PreToolUse hook for the Read tool set -euo pipefail # Read the hook input (tool name and arguments) input=$(cat) # Parse the file_path from the Read tool arguments file_path=$(echo "$input" | jq -r '.arguments.file_path // empty') # If no file_path, allow the read to continue if [[ -z "$file_path" ]]; then echo '{"continue": true}' exit 0 fi # Check if file exists if [[ ! -f "$file_path" ]]; then # File doesn't exist yet - let Read tool handle the error echo '{"continue": true}' exit 0 fi # Get file size in bytes file_size=$(stat -f%z "$file_path" 2>/dev/null || stat -c%s "$file_path" 2>/dev/null || echo "0") # Threshold: 10KB = 10000 bytes threshold=10000 # If file is larger than threshold, suggest RLM tools if [[ "$file_size" -gt "$threshold" ]]; then # Format file size for display if [[ "$file_size" -gt 1048576 ]]; then size_display="$(echo "scale=2; $file_size / 1048576" | bc)MB" elif [[ "$file_size" -gt 1024 ]]; then size_display="$(echo "scale=2; $file_size / 1024" | bc)KB" else size_display="${file_size} bytes" fi message="📊 Large file detected: $file_path ($size_display) Consider using RLM tools for efficient processing: 1. rlm_load_context: Load file externally (keeps context clean) 2. rlm_chunk_context: Break into manageable pieces 3. rlm_sub_query_batch: Process chunks in parallel with sub-LLM calls This approach prevents context bloat and enables processing files that would otherwise exceed token limits." # Return JSON with message but allow read to continue jq -n \ --arg msg "$message" \ '{ "continue": true, "message": $msg }' else # File is small enough - proceed normally echo '{"continue": true}' fi