Claude setup¶
This page is only about using the MCP server from Claude. For the general memory rules that apply to any AI tool, start with the AI memory playbook.
Which file should I use?¶
| File | Purpose |
|---|---|
| AI memory playbook | Recommended operating pattern for any LLM |
Sample CLAUDE.md |
Copyable Claude project-instructions sample |
| This page | Claude-specific setup pointer |
Add the memory policy¶
After registering the liel MCP server with Claude, add a project memory
policy to your Claude project instructions.
For a ready-to-adapt sample, see sample CLAUDE.md.
The most important rule is:
Always check existing memory before asking the user to repeat context.
Why this matters¶
Claude will only use a memory tool well if the project instructions define both read discipline and write discipline:
- read memory before asking the user to repeat context
- save only durable, high-signal information
- write at meaningful checkpoints, not every turn
- use nodes for entities and edges for relationships
Those rules are explained in the AI memory playbook.