← Back to Palim

Claude Code persistent memory

Give Claude Code persistent memory across sessions

Claude Code can read durable project instructions from files such as CLAUDE.md. That is ideal for stable rules, but not for every decision, failed approach, or debugging discovery produced during daily work.

Palim adds a hosted MCP memory alongside those files. You explicitly save the sessions and decisions worth keeping, then retrieve them with full-text search from Claude Code or another connected MCP client.

Set up Palim for Claude Code

What this workflow preserves

  • Architecture decisions and the reasoning attached to them
  • Debugging discoveries you do not want to repeat
  • Session summaries, tags, timestamps, and source context
  • Context that should also be available in Cursor, Claude, or ChatGPT

Set up persistent memory in Claude Code

  1. 1

    Connect the remote MCP server

    Add Palim as an HTTP MCP server. Claude Code opens the OAuth flow when it first needs the connection.

    claude mcp add --transport http palim https://api.usepalim.com/mcp
  2. 2

    Keep durable rules in CLAUDE.md

    Use CLAUDE.md for short, stable project instructions. Add a rule telling Claude when to save important sessions and when to search Palim.

  3. 3

    Save work worth carrying forward

    At the end of a useful session, ask Claude Code to save it. Palim stores only what you explicitly ask it to store.

    Save this session to Palim with the decisions, failed approaches, and next steps.
  4. 4

    Recall it in a later session

    Search with concrete terms from the earlier work. Palim currently uses full-text and exact text matching, not semantic embeddings.

    Search Palim for the authentication redirect decision and summarize what we chose.

CLAUDE.md and MCP memory solve different problems

CLAUDE.md is versioned project context: coding conventions, commands, architectural constraints, and instructions that should be loaded repeatedly. Palim is a searchable record of work that emerged over time.

Use both. Keep the small set of durable rules in the repository and save evolving sessions, decisions, and discoveries to Palim. This avoids turning one instruction file into an unmaintainable project diary.

Memory that is not locked to Claude Code

A decision saved from Claude Code can be retrieved from Cursor, Claude, ChatGPT, or another connected MCP client. The shared layer matters when the next task starts in a different tool or on another device.

  • One hosted MCP endpoint
  • OAuth for supported clients
  • AES-256-GCM encryption for supported stored content fields
  • EU database region in Frankfurt

Frequently asked questions

Does Claude Code already have persistent memory?

Claude Code supports persistent instructions and memory files. Palim complements them with an explicit, searchable session and decision archive that can also be used from other MCP clients.

Does Palim save every Claude Code conversation automatically?

No. Palim saves a session or memory when you or the connected client explicitly call the relevant Palim tool.

Is Palim semantic or vector search?

Not currently. Search uses PostgreSQL full-text search and text matching, so concrete keywords, tags, and varied query terms work best.

Can Cursor retrieve a decision saved from Claude Code?

Yes. Connect both clients to the same Palim account, save the decision in Claude Code, and search for it from Cursor.

Start with one decision you do not want to lose

Connect Claude Code, save one real session, and retrieve it from a fresh chat.

Set up Palim for Claude Code