All notable changes to the MCP Claude Code project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Added new USEFUL_PROMPTS.md document with practical prompt templates for common scenarios:
- Summarizing history for continued conversations
- Automation research summaries
- Release preparation commands
- Resuming interrupted conversations
- Reduced default values for agent tool limits:
- agent_max_iterations reduced from 30 to 10
- agent_max_tool_uses reduced from 100 to 30
- Enhanced system prompt documentation for the dispatch_agent tool:
- Improved guidelines for when to use the dispatch_agent tool
- Added detailed examples for effective agent prompting
- Clarified agent limitations and capabilities
- Added categorized use case recommendations
- Improved README.md with clearer feature descriptions
- Fixed formatting and alignment in README.md tables
- Added cross-reference to USEFUL_PROMPTS.md in the README
- Added specific guidance for file reading operations in system prompt
- Added Agent Tool for task delegation and concurrent execution
- Agent Tool enables Claude to delegate complex tasks to specialized sub-agents that can work in parallel, enhancing performance for complex operations like code search, analysis, and multi-step tasks
- Agents use read-only tools to ensure safety while maintaining full access to information retrieval capabilities
- Introduced new parameter
--enable-agent-toolto allow users to control whether the agent tool is registered - Added parameters
--agent-max-iterations(default: 30) and--agent-max-tool-uses(default: 100) to control agent behavior - Implemented multi-agent support to enable concurrent sub-agent execution for improved performance when handling multiple related tasks
- Agents return results that are automatically formatted and integrated into Claude's responses
- Added LiteLLM integration to support various model providers beyond OpenAI
- Command line support for specifying the model name, API key, and token limits
- Support for Anthropic, OpenAI, Google, and other LLM providers
- Simplified configuration through environment variables or command-line arguments
- Enhanced system prompt with best practices and dispatch agent usage guidelines
- Improved agent tool implementation with better parameter validation and error handling
- Increased max_tool_uses limit from 15 to 30 for improved agent flexibility
- Optimized agent implementation to avoid redundant checks and improve performance
- Restricted agent tool to use only read-only variants of filesystem tools for security
- Simplified agent tool implementation to focus on multi-agent operation
- Refactored system prompt handling and improved user prompt support
- Updated parameter validation to support both string and array inputs for prompts
- Fixed JSON serialization error in agent tool error handling
- Fixed redundant type checks and improved test environment handling
- Improved path description in read_files schema for better clarity
- Updated type annotations to use built-in generics syntax
- Added detailed documentation for the agent tool in INSTALL.md, explaining setup and configuration options
- Added debugging guide for Model Context Protocol Inspector
- Updated installation instructions with LLM model and token configuration options
- Clarified cwd parameter restrictions in shell tool documentation
- Significantly improved performance of
search_contentandcontent_replacetools - Implemented parallel processing for file searching with batched execution
- Optimized file finding strategy using more efficient directory traversal
- Added semaphore-based concurrency control to avoid overwhelming the system
- Replaced recursive directory traversal with more efficient
pathlib.Path.rglob()method - Restructured file search logic to filter allowed paths more efficiently
- Changed file pattern matching to use
fnmatchfor more consistent behavior
- Enhanced
search_contenttool to use regular expression pattern matching instead of simple substring matching - Added comprehensive test suite for regex search functionality
- Improved code search capabilities with pattern matching support
- Modified search implementation to use
re.search()instead of theinoperator - Maintained backward compatibility with existing string search patterns
- Added Jupyter notebook support with
read_notebookandedit_notebooktools - Implemented reading of notebook cells with their outputs (text, error messages, etc.)
- Added capabilities for editing, inserting, and deleting cells in Jupyter notebooks
- Added comprehensive test suite for the notebook operations
- Updated tools registration to include the new Jupyter notebook tools
- Enhanced README.md with Jupyter notebook functionality documentation
- Added support for file paths in
search_contentandcontent_replacetools - Extended functionality to search within and modify a single file directly
- Implemented smarter path handling to detect file vs directory inputs
- Updated docstrings to clarify that
pathparameter now accepts both file and directory paths - Enhanced tool parameter descriptions for improved clarity
- Added extensive test coverage for the new file path functionality
- Improved error messaging for file operations
- Enhanced
directory_treetool with depth limits and filtering capabilities - Added parameter
depthto control traversal depth (default: 3, 0 or -1 for unlimited) - Added parameter
include_filteredto optionally include commonly filtered directories - Added statistics summary to directory tree output
- Improved directory tree output format from JSON to more readable indented text
- Added filtering for common development directories (.git, node_modules, etc.)
- Enhanced directory tree structure to show skipped directories with reason
- Improved README instructions regarding the placement of the system prompt in Claude Desktop
- Clarified that the system prompt must be placed in the "Project instructions" section for optimal performance
- Modified permissions system to allow access to
.gitfolders by default - Updated tests to reflect the new permission behavior
- Improved documentation in file_operations.py to clarify that the
pathparameter refers to an absolute path rather than a relative path - Enhanced developer experience by providing clearer API documentation for FileOperations class methods
- Enhanced release workflow to reliably extract and display release notes
- Improved error handling and fallback mechanism for the release process
- Fixed GitHub Actions workflow to properly display release notes from CHANGELOG.md
- Added fallback mechanism when release notes aren't found in CHANGELOG.md
- Added "think" tool based on Anthropic's research to enhance Claude's complex reasoning abilities
- Updated documentation with guidance on when and how to use the think tool
- Added validation in
edit_fileto ensureoldTextparameter is not empty
- Fixed GitHub Actions workflow permissions for creating releases
- Updated GitHub Actions to latest versions (v3 to v4 for artifacts, v3 to v4 for checkout, v4 to v5 for setup-python)
- Added UVX support for zero-install usage
- Simplified README.md to focus only on configuration with uvx usage
- Updated command arguments in documentation for improved clarity
- Fixed package structure to include all subpackages
- Updated build configuration to properly include all modules
- Published to PyPI for easier installation
- Improved package metadata
- Initial public release
- Complete MCP server implementation with Claude Code capabilities
- Tools for code understanding, modification, and analysis
- Security features for safe file operations
- Comprehensive test suite
- Documentation in README
- Improved error handling in file operations
- Enhanced permission validation
- Version synchronization between package files
- Initial development version
- Basic MCP server structure
- Core tool implementations