Docker Manager MCP

# Tool Token Consumption Tracking ## Baseline (Before Field Description Enhancements) Track token consumption for our Docker MCP tools before and after adding enhanced Field descriptions with behavioral hints. ### Initial Token Usage (With Annotations Only) - **mcp__docker__docker_hosts**: 1.9k tokens - **mcp__docker__docker_container**: 865 tokens - **mcp__docker__docker_compose**: 1.1k tokens **Total**: 3,865 tokens ### Annotations Added - User-friendly titles - readOnlyHint, destructiveHint, idempotentHint, openWorldHint flags - These don't consume token context in LLM prompts --- ## After Field Description Enhancements (To Be Measured) Will track token consumption after adding: 1. Behavioral hints in action descriptions 2. Safety warnings for critical parameters 3. Guidance for safe/recommended defaults 4. Examples for complex parameters 5. Enhanced validation constraints ### Expected Changes - Field descriptions ARE sent to LLM (consume tokens) - More descriptive text = higher token usage - But provides better context for LLM decision making ### Post-Enhancement Token Usage #### After Field Description Enhancements: - **mcp__docker__docker_hosts**: 1.9k tokens (unchanged) - **mcp__docker__docker_container**: 865 tokens (unchanged) - **mcp__docker__docker_compose**: 1.3k tokens (+200 tokens, +18.2%) **Total**: 4,065 tokens (+200 tokens, +5.2% overall) #### After Docstring Enhancement (bullet points + optional params): - **mcp__docker__docker_hosts**: 1.9k tokens (unchanged) - **mcp__docker__docker_container**: 865 tokens (unchanged) - **mcp__docker__docker_compose**: 1.4k tokens (+100 tokens more, +300 total, +27.3%) **Final Total**: 4,165 tokens **Total Increase**: +300 tokens (+7.8% overall, +27.3% for docker_compose) #### After Enum & Validation Constraints Enhancement: Changes implemented: 1. **Enum Classes**: Replaced `Literal` types with proper Enum classes (HostAction, ContainerAction, ComposeAction, Protocol) 2. **Pattern Validation**: Added DNS-compliant pattern for stack_name (no underscores) 3. **Length Constraints**: Added min_length (26) for compose_content, max_length for string fields 4. **Numeric Constraints**: Enhanced existing ge/le constraints for ports and other numeric fields 5. **Implementation Fix**: Updated stack_name validation in server.py to match DNS compliance **Final Token Usage Estimates:** - **mcp__docker__docker_hosts**: ~2.3k tokens (+400 tokens, +21.1%) - **mcp__docker__docker_container**: ~1.1k tokens (+235 tokens, +27.2%) - **mcp__docker__docker_compose**: ~1.6k tokens (+500 tokens, +45.5%) **Final Total**: ~5,000 tokens (+1,135 tokens total, +29.4% overall increase from baseline) **Enhancement Summary:** - Started at 3,865 tokens (baseline) - Ended at ~5,000 tokens - Total increase: ~1,135 tokens (+29.4% overall) - Most impact on docker_compose tool (+45.5% from baseline) #### After Docstring Formatting Enhancement: **Final Changes:** 6. **Docstring Formatting**: Updated docker_hosts and docker_container docstrings to match docker_compose bullet-point style - Better organization with • bullets and - sub-bullets - Clear separation of required vs optional parameters - Consistent formatting across all 3 tools --- ## Notes - Annotations (readOnlyHint, etc.) don't consume tokens - they're metadata for client apps - Field descriptions are part of tool schema sent to LLM - they DO consume tokens - Enum types provide better type safety and IDE support with minimal token overhead - Pattern validation prevents invalid inputs at the parameter level - Trade-off: More tokens for better LLM understanding, type safety, and safer operations