Published to PyPI, Official MCP Registry, and Smithery.ai
MCP Server mode: use as a Model Context Protocol server for programmatic optimization
Available via pip install mcp-credit-optimizer
Enhanced documentation with installation guides for all platforms
Expanded audit to 53 adversarial scenarios (up from 22)
Added 3 new optimization techniques: Output Scoping, Parallel Decomposition, Incremental Refinement
Improved Smart Testing detection for code-heavy tasks
Better handling of mixed-intent prompts (research + code + writing)
Complete rewrite of the optimization engine with 10 techniques
Intelligent model routing: Standard vs Max based on task complexity
Context hygiene system to reduce token waste
Section-by-section processing for long-form content
Factual data detection to prevent hallucination in data-heavy tasks
Zero quality loss guarantee — audited across 22 scenarios
30-75% average credit savings (55% typical)
Initial public release as Manus Skill
Basic model routing (Standard vs Max)
Simple prompt analysis for task classification
~30% average credit savings
Model Routing
Standard vs Max based on complexity
Smart Testing
Detect when testing is unnecessary
Output Scoping
Limit output to what's needed
Context Hygiene
Reduce token waste in prompts
Parallel Decomposition
Split tasks for efficiency
Incremental Refinement
Build iteratively, not all at once
Chat Mode Routing
Route simple Q&A to free tier
Section Processing
Handle long content in parts
Factual Detection
Prevent hallucination in data tasks
Mixed Intent Analysis
Optimize multi-type prompts