- Added missing tool descriptions to generate_standup_report, get_environment_details, and list_context_workspaces so the LLM correctly parses them during MCP handshake.
- Removed heavy Tantivy index rebuild loop that ran every 5 seconds and rewrote entire graph to disk
- Converted polling worker to lightweight IndexWriter committer
- Fixed 67+ instances of unnecessary clone and string parsing overhead in tool return results
- Deduplicated state graph modification methods
Created a parse_tool! macro that eliminated ~450 lines of repetitive error-handling boilerplate across 54 MCP tool definitions in server/src/handlers.rs.
Purged 30+ stale, redundant python patching scripts and legacy mcp_client.py test runners that were polluting the workspace root and server/src directories.
- Added native E2E test suite in stub to validate server<->stub<->nvim IPC
- Added win-nvim integration tests
- Fixed empty capabilities schema in initialize response in server and win-nvim
- Refactored server to accept dynamic MCP_PORT for test isolation
- Made server exit gracefully on port 3000 collision
Makes apply_sync_write, write_to_local_delta, and store modify truly async, preventing the mcp-memory-server from locking the tokio executor during disk IO
Riz Ashraf