research-server-patterns

Recipe: Memory Persistence for Claude Code

The problem: You explain a codebase to Claude on Monday. Friday, it has no idea what you were working on.

This recipe: Claude remembers Monday on Friday. No re-explaining. No losing the thread mid-project.

What it does

Adds 5 tools to every Claude Code session:

memory_save — persist a decision, pattern, or finding
memory_read — retrieve it by key
memory_search — full-text search across everything saved
session_save — save a structured end-of-session summary with next steps
session_get — load last session’s context at the start of a new one

Memory persists in a local SQLite file. Nothing leaves your machine.

Setup (5 minutes)

Requirements: Node.js 18+

1. Clone and install

git clone https://github.com/shukumei7/research-server-patterns
cd research-server-patterns/recipes/memory-persistence
npm install

2. Add to Claude Code

claude mcp add research-memory node /absolute/path/to/recipes/memory-persistence/mcp-memory.js

Replace /absolute/path/to/ with the actual path on your machine.

Windows example:

claude mcp add research-memory node "D:\research-server-patterns\recipes\memory-persistence\mcp-memory.js"

3. That’s it. Start a Claude Code session. The memory_save, memory_read, memory_search, session_save, and session_get tools are now available.

The Monday-to-Friday workflow

Monday: At the end of your session, tell Claude:

“Save a session summary for this project.”

Claude calls session_save with what was accomplished and next steps.

Friday: At the start of your new session, tell Claude:

“Load last session for my-project.”

Claude calls session_get and picks up exactly where you left off — decisions made, context intact, next steps ready.

Any time: Save a specific decision so it survives context compaction:

“Remember that we’re using JWT with 24h expiry for auth — save that as a memory.”

Claude calls memory_save. Every future session in that project can retrieve it.

Example: saving context that survives compaction

You: We decided to use edge caching for the API responses. Save that.
Claude: [calls memory_save: key="caching-decision", content="Use edge caching for all GET /api/* routes. TTL=300s. Invalidate on write."]

[two weeks later, fresh session]

You: What did we decide about caching?
Claude: [calls memory_search: query="caching"]
Result: "Use edge caching for all GET /api/* routes. TTL=300s. Invalidate on write."

Custom database path

Set MEMORY_DB_PATH to an absolute path to control where the SQLite file lives:

MEMORY_DB_PATH=/Users/you/memory/projects.db claude

Or add to your shell profile to make it permanent.

vs. cloud memory alternatives

If you want semantic search and cross-device sync, Mem0 is the cloud-hosted alternative with a free tier. This recipe trades those features for:

No account required — nothing to sign up for, no third-party memory service
Local storage — memory lives in a SQLite file on your machine; no data goes to a memory cloud service. Note: memory content is passed to Claude through your normal Claude Code sessions, which use Anthropic’s API.
Works offline — no memory service network dependency (Claude Code itself still needs API access)
Auditable — open the .db file with any SQLite browser and read every memory directly

This is one recipe from Research Server

Research Server is a full autonomous back-office system for solo founders — email triggers, scheduled agents, SR&ED diary automation, and more. This recipe is the memory layer, standalone.

→ Research Server on GitHub
→ Done-for-you setup ($2,500)