Silent knowledge persistence for Claude Code — your AI pair programmer remembers what you've learned.
Engram automatically captures problem-solving knowledge from your Claude Code sessions and resurfaces it when you encounter similar situations. No manual note-taking, no copy-pasting into files — it just works.
You solve a tricky problem with Claude Code. A week later, you hit the same situation. But the session is gone, your memory is fuzzy, and you start from scratch.
- Auto-surface: When Claude sees the engram tools, it checks the keyword index and pulls in relevant knowledge on demand.
- Auto-save: When you solve a non-trivial problem, Claude saves the learning — what the problem was, what you tried, what worked, and the final solution.
- Cross-session: A shared local SQLite database means knowledge from any session is available to all others.
Each captured entry contains:
| Field | Description |
|---|---|
| Situation | What was the context? |
| Tough spot | What was the challenge? |
| Approach | How did we tackle it? |
| Outcome | What worked and what didn't? |
| Solution | Final resolution |
| Keywords | For retrieval and matching |
# 1. Clone and install
git clone https://github.com/zzzhouzhenzz/engram.git
cd engram
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
# 2. Register MCP server (run once)
claude mcp add --scope user engram_mcp_server -- $(which engram_mcp_server)Claude Code manages the server lifecycle automatically via stdio — no manual server management needed.
| Tool | Description |
|---|---|
get_keyword_index |
List all keywords in the knowledge base |
query_knowledge |
Search by keywords |
get_recent_knowledge |
Get N most recent entries |
save_knowledge |
Save a learning from the current session |
MIT