Your AI Coding Agent Forgets Everything. I Fixed It With a Local MCP Memory Server

说明与用途

本地 MCP 内存服务器让 AI 编码助手在会话之间保留上下文，消除重复解释代码和决策的需求。它使用 Markdown 文件和 SQLite FTS5 进行离线搜索，Recall@10 达到 98.6%，无需 LLM 调用。

问题

AI coding agents lack persistent memory across sessions, forcing developers to repeatedly explain their codebase and project decisions.

结果

The local MCP memory server achieves Recall@1 80.2%, Recall@3 92.8%, Recall@5 96.0%, Recall@10 98.6% on LongMemEval, outperforming single BM25 or vector methods by 3% with zero LLM calls on retrieval.

工作流

• Run npx @awareness.market/setup to install and start the local daemon on 127.0.0.1:37800.
• The installer detects the IDE and wires the MCP server.
• The agent gains five tools: awareness_init, awareness_recall, awareness_record, awareness_lookup, awareness_get_agent_prompt.
• During a session, the agent calls awareness_init to load recent context.
• When needing information, the agent uses awareness_recall in two phases: first a summary, then full content for selected items.
• The agent records new decisions with awareness_record.
• The memory is persisted in .awareness/ as Markdown files and an SQLite FTS5 index.
• Retrieval uses hybrid BM25 + vector embeddings with RRF, no LLM calls.
• The agent can recall past migrations, decisions, and TODOs automatically.
• The process is fully offline and local-first.

证据

I evaluated it on LongMemEval (ICLR 2025) — 500 human‑curated questions, ~115k tokens of history per question: Recall@1 80.2% Recall@3 92.8% Recall@5 96.0% ◀ primary Recall@10 98.6%

来源

Retrieval code and scripts are in the repo, so you can re‑run it yourself rather than trust a table in a blog post.

来源

Memory lives in .awareness/, as Markdown you can read, edit, diff, and commit: .awareness/ ├── memories/2026-03-22_decided-to-use-postgresql.md ├── knowledge/decisions/postgresql-over-mysql.md ├── tasks/open/implement-rate-limiting.md └── index.db # SQLite FTS5 index, rebuilt automatically

来源