All notable changes to MCPMan will be documented in this file.

• Updated dependency versions
• Fixed tool shape differences between openai_compat and openai_responses

• Improved reasoner detection for better tool orchestration

• Added meta-MCP-agent-agent capability for enhanced tool orchestration
• Improved environment argument handling for better flexibility

• Fixed TypeError in logger when response_time is None
• Fixed formatting issue with logging HTTP and LLM responses

• Improved CLI error messages for missing required arguments
• Enhanced error output to be more helpful to users

• Fixed issue with missing arguments validation in CLI

• Significantly improved error handling throughout the entire codebase
• Added fallback server capability for increased resilience
• Added robust message parsing to handle malformed data

• Fixed issue where errors in tool calls would terminate the entire process
• Fixed error handling in LLM client communications
• Fixed error propagation to allow LLMs to self-heal from errors
• Improved error messages for users

• Added enhanced structured logging system with run ID tracking
• Added mcpreplay tool for visualizing previous conversations from log files
• Added replay mode to CLI with --replay and --replay-file parameters
• Added standardized logging across all LLM clients

• Added OpenAI (v1.75.0) and Anthropic (v0.49.0) client libraries as explicit dependencies
• Improved dependency management for better user experience

• Fixed critical bug in array schema validation for OpenAI tools without items type
• Fixed incorrect handling of tools with array parameters that didn't specify item types

• Added colorama dependency explicitly to requirements
• Improved compatibility with Claude 3.5 Sonnet model
• Added standardized content handling across all models
• Added proper support for Anthropic's Claude models
• Added documentation for Claude model support in README

• Fixed missing dependency handling in packaging
• Fixed issue with Claude repeatedly calling the same tool by implementing intelligent deduplication that handles:
  • Removing exact duplicate tool calls (same name and arguments)
  • Allowing multiple calls to the same tool with different arguments
  • Properly handling JSON object key ordering in arguments comparison
• Fixed handling of messages without content in Anthropic client

• Refactored codebase to use a domain model approach with cleaner abstractions
• Simplified orchestrator.py with better separation of concerns
• Improved error handling and logging

• Improved max_turns handling with better help text

• Fixed version numbering

• Fixed issues found in end-to-end tests