Skip to content

Conversation

partouf
Copy link
Member

@partouf partouf commented Oct 16, 2025

Summary

  • Reduced tool docstring verbosity from ~2000 tokens to ~500 tokens per tool
  • Addresses issue Reduce token count #4 by matching token efficiency of other MCP implementations (e.g., GitHub's 500-600 tokens)
  • All 12 MCP tools updated with concise documentation

Changes

  • Removed verbose "Use Cases" sections
  • Condensed parameter descriptions to inline format
  • Eliminated extensive example sections
  • Removed "When to use vs other tools" comparison sections
  • Preserved essential information: parameters, return values, and core functionality

Testing

⚠️ Note: While unit tests pass (81/81), the actual MCP behavior with Claude has not been tested to confirm tools still function
correctly with the reduced docstrings.

Token Reduction Impact

Tool Before After
compile_check_tool ~1500 tokens ~400 tokens
compile_and_run_tool ~1800 tokens ~450 tokens
compile_with_diagnostics_tool ~1900 tokens ~400 tokens
analyze_optimization_tool ~2300 tokens ~450 tokens
compare_compilers_tool ~795 tokens ~350 tokens
generate_share_url_tool ~2100 tokens ~400 tokens
find_compilers_tool ~1500 tokens ~600 tokens
Other tools 700-1000 tokens 300-500 tokens

Related

Fixes #4

Reduced verbosity from ~2000 to ~500 tokens per tool to match other MCP
implementations. All 12 tool docstrings have been made more concise while
preserving essential parameter and return value information.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>
@partouf
Copy link
Member Author

partouf commented Oct 16, 2025

This is missing way too much information

- Preserved essential use cases and tool differentiation guidance
- Kept critical warnings (e.g., find_compilers token limits)
- Maintained assembly analysis patterns for optimization tool
- Added concise examples showing proper parameter format
- Preserved 'when to use vs other tools' guidance
- Target: ~800-1000 tokens per tool (vs original 2000+)

All tests pass (81/81). Addresses feedback on initial reduction being too aggressive.
Use more natural 'simplified' instead of technical AI-focused language.
@partouf
Copy link
Member Author

partouf commented Oct 16, 2025

possibly better now, still havent tested

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Reduce token count

1 participant