Refactored to unified npm package (agentic-mcp)
Replaced HTTP with Socket communication (newline-delimited JSON protocol)
Merged cli/ and daemon/ into single src/ structure
Archived Python scripts to archive/python-legacy/
Updated all commands from mcp to agentic-mcp
Updated system architecture diagrams (Socket instead of HTTP)
Updated Environment Variables (removed MCP_DAEMON_PORT)
Changed license to Apache-2.0 with acknowledgments
Updated README.md and README_zhTW.md for current implementation
Added custom description field support in mcp-servers.json for AI context
Added description?: string to MCPServerConfig type definition
Added getCachedMetadata() method to ProgressiveMCPClient for daemon access
Added GET /metadata?sessionId=xxx endpoint to return actual MCP server metadata
Fixed Layer 1 metadata retrieval to return name, version, capabilities, and description
Fixed config file path resolution to read mcp-servers.json from project root
Added MCP_DAEMON_CONFIG environment variable support for custom config path
Updated mcp_metadata.py to use new /metadata endpoint instead of tools/list
Updated daemon_start.py to calculate correct project root path and set MCP_DAEMON_CONFIG
Updated loadServerConfigs() to prioritize MCP_DAEMON_CONFIG > root config > daemon config
Updated README.md with description field examples and config file location priority
Updated README_zhTW.md with description field examples and config file location priority
Created .claude/context/project.md with comprehensive project documentation
Created CHANGELOG.md for tracking changes
Updated mcp-servers.json with English description for playwright server
Initial proof-of-concept release
Three-layer progressive disclosure (metadata → tools → schema)
Daemon architecture with persistent MCP connections
Support for stdio, http-streamable, and SSE transports
Global and dynamic session management
Hot reload configuration support
You can’t perform that action at this time.