Claude Code Memory

A Claude Code plugin that turns your own session history into a persistent, queryable knowledge graph — a second brain across every repo you work in — visualized natively in Obsidian.

The problem this solves

If you work across multiple repos from Claude Code, every fresh session starts from zero: you re-explain architecture, re-state conventions, and re-grant permissions you already gave last week. This plugin captures that once and makes it available automatically going forward.

How it works

• A vault — a plain folder of markdown notes (default ~/ClaudeMemory-vault), wikilinked by entity (Project, Convention, Decision, Permission, Preference, Session, and domain-specific notes like Schema). Open it as an Obsidian vault and the built-in Graph view renders it, no extra plugins needed.
• An MCP server (mcp-server/) — runs automatically once this plugin is enabled, giving Claude tools to read from and write to the vault:
  • get_project_context(repo_path) — Claude calls this before starting work in a repo, pulling in everything remembered about it (architecture, conventions, decisions, permissions, schema) without you re-explaining anything.
  • search_memory, get_entity, list_related — deeper on-demand queries.
  • remember(...) — Claude calls this live, mid-conversation, whenever you tell it something worth keeping.
• A sync-memory skill — invoked manually as /sync-memory <days>, it reads your actual Claude Code transcripts (~/.claude/projects/**/*.jsonl) for the given day window and extracts them into the vault. Small deltas are processed serially; a large backfill (first run, or a long gap) fans out across subagents — one per project/date-range batch, each purely extracting and reporting back — with a single reconciliation pass doing all the actual writing, so parallel extraction never races on the same file.

Setup

1. Enable this plugin in Claude Code (adds the memory MCP server and the sync-memory skill).
2. Optionally set CLAUDE_MEMORY_VAULT_DIR if you don't want the default ~/ClaudeMemory-vault location.
3. Run /sync-memory 30 (or however far back you want) to do a first import.
4. Open the vault folder in Obsidian and look at Graph view.

From then on, get_project_context and remember work automatically in every session; run /sync-memory <days> again whenever you want to pull in more history.

Repo layout

.claude-plugin/plugin.json   plugin manifest
.mcp.json                    registers the memory MCP server
mcp-server/                  Node stdio MCP server (read + write tools over the vault)
skills/sync-memory/          the /sync-memory <days> skill and its recon/state scripts
vault-template/              seeds a new vault: Obsidian graph color config + a starter note