Integration: Claude Code

Integrating l0-memory with Claude Code allows your terminal-based assistant to save and retrieve memories directly while you work.

Automatic Installation

If you have the source code cloned, you can use the provided Makefile to quickly add the server to Claude Code:

make install-mcp

This command runs claude mcp add with the correct path to your local ltm binary.

Manual Installation

To manually add l0-memory to Claude Code, use the following command:

claude mcp add l0-memory /absolute/path/to/ltm mcp

Note

Replace /absolute/path/to/ltm with the actual path where you installed the binary. If it's in your PATH, you can just use ltm.

Verification

Once added, you can verify that Claude Code recognizes the new tools:

1. Start Claude Code: claude
2. Ask it to list its tools: "What MCP tools do you have access to?"
3. You should see tools prefixed with memory_ (e.g., memory_save, memory_get).

Usage in Claude Code

Claude Code will automatically use these tools when it identifies information worth remembering. You can also explicitly command it:

"Remember that this project uses a hexagonal architecture."

The assistant will then call memory_save to persist this fact in the current scope.

Scope Awareness

Claude Code is generally aware of your current directory. It will often default to a repo: scope if it can identify the repository name, keeping your project memories organized.

Automatic recall + /checkpoint

The MCP server above lets the assistant read and write memories on request. The optional Claude Code integration (integrations/claude-code/) makes it automatic — so a session resumes without you re-explaining who you are or where the project left off:

Piece	What it does	Trigger
Recall hook	A SessionStart hook that injects your persona (pinned user-scope entries) and the current project memory (repo:<slug>, pinned first) into the session.	Automatic, every session
/checkpoint skill	Synthesizes a durable state snapshot and saves it to the right scope, pinning the durable one.	You run /checkpoint

Install both with:

make install-claude
# or: integrations/claude-code/install.sh

This copies the hook and skill into ~/.claude/ and registers the hook in ~/.claude/settings.json (idempotent). Open a new session to pick it up.

You own the hook

Registering a SessionStart hook makes a script run on every session start. The installer is something you run, so you own that change — review integrations/claude-code/l0-recall.py first if you like. The hook reads via the ltm CLI (pinned / list are pure SQLite reads), truncates each entry, and fails open (any error injects nothing and never blocks a session).

The loop

1. Open a repo → the hook injects "who you are" + "this project's state".
2. Work.
3. /checkpoint at the end → it saves the updated state, pinning the durable entry.
4. Next session — even from another MCP host — starts at step 1, context already loaded.

Pinning is the relevance lever

The recall hook surfaces pinned entries first and falls back to recent ones. Pin what you always want on open; leave transient working notes unpinned.

ltm --scope user        pin identity       # persona facts, surfaced everywhere
ltm --scope repo:myproj pin release-state  # the canonical project state

See integrations/claude-code/README.md for configuration (LTM_BIN, L0_RECALL_MAXVALUE, L0_RECALL_MAXREPO).