prefrontal-systems/mcp-smart-reader
If you are the rightful owner of mcp-smart-reader and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.
MCP Smart Reader is a server that efficiently processes large documents by summarizing them, saving tokens and improving context window efficiency.
MCP Smart Reader
Smart document reader MCP server with automatic summarization for large files.
Overview
MCP Smart Reader provides token-efficient access to large documents by automatically summarizing files that exceed a token threshold. Instead of loading a 50K token document into Claude's context every message, it returns an 800 token summary with rich metadata, dramatically improving context window efficiency.
Key Features
- Automatic Summarization: Files >10K tokens are automatically summarized
- Rich Metadata: Summaries include coverage info, suggestions, and cross-references
- Section Extraction: Read specific sections without loading entire document
- Token Tracking: Know exactly how many tokens you're saving
- Actionable Guidance: Claude knows when to request more detail
Token Savings Example
Without smart_read:
paper.md: 53,132 tokens every message
3 exchanges: 159,396 tokens consumed
With smart_read:
Summary: 797 tokens
3 exchanges: 2,391 tokens consumed
Savings: 157,005 tokens (98.5% reduction)
Installation
# Clone repository
cd mcp-smart-reader
# Install with uv (recommended)
uv sync
# Or with pip
pip install -e .
Development Status
Current: Basic structure in place, core functionality implemented
Next Steps:
- Implement section extraction logic
- Add multi-format support (PDF, DOCX)
- Enhance summarization algorithms
- Add caching layer
- Comprehensive testing
License
Apache 2.0
Author
Scot Campbell
Prefrontal Systems LLC
scot@prefrontal.systems