mirror of
https://github.com/smart-memory/smart-memory-mcp.git
synced 2026-08-21 15:43:29 +00:00
smartmem-dev
1fd2ab388c
The cross-encoder reranker loads lazily and returns fusion order UNRANKED until warm (core DIST-LITE-WARMSTART-1). Only a search triggered that load, so the first transcript_search of every session was unranked. Measured on the 386-chunk corpus, same process, same query: call 1 put /unflush boilerplate at ranks 1-2 with the correct session outside the top 6; call 2 returned it at rank 1. server.main() now schedules core's existing warm hook (warm_search_models_async, CORE-SEARCH-WARMSTART-1 — the same one service.py runs in its lifespan). Gated on a transcript store existing (this module is imported for every Claude Code session; transcripts are opt-in) and on SMARTMEMORY_WARM_RERANKER, matching the service flag exactly. Deliberately NOT in register() — the test suite drives that, and it would load a real model whenever the developer has a store on disk — and not in _get_memory(), which runs inside the first search and would turn a certain miss into a race. Verified end to end against the real store: first search after boot returns the correct session at rank 1, boot still returns in 0.02s. 39 tests pass.
1fd2ab388c
·
2026-08-20 13:11:03 +08:00
109 commits
smart-memory-mcp
Unified SmartMemory MCP (Model Context Protocol) server — tiered tools, local + remote backends.
Overview
MCP server exposing SmartMemory operations to MCP-compatible clients (Claude Desktop, Cursor, etc.). Implements the full memory toolset (add, search, recall, decisions, plans, anchors, code-index) and routes to either a local SmartMemory instance or a remote smart-memory-service API endpoint.
Status
Version: 0.2.1
Quick start
pip install -e .
# Run the server
smartmemory-mcp # or python -m smartmemory_mcp
Tests:
pytest tests/ -v
Documentation
Full SmartMemory documentation: https://docs.smartmemory.ai
Part of SmartMemory
This is one component of the SmartMemory ecosystem. See the main repo for the broader project.