Persistent Memory for Claude

Claude is one of the most capable AI assistants available, but it suffers from a critical limitation: every conversation starts from scratch. Whether you are using Claude Desktop for daily tasks, Claude Code for software development, or the Anthropic API for building applications, Claude has no built-in mechanism to remember what you discussed yesterday, what your project architecture looks like, or what coding conventions your team follows. PersistMemory fixes this by giving Claude a persistent, searchable memory layer through the Model Context Protocol that Anthropic itself created.

Why Claude Forgets Everything Between Sessions

Claude operates as a stateless model. Each API call receives a sequence of tokens and produces a response, with no internal state carried between calls. The conversational flow you experience in Claude Desktop is maintained by the application resending the full conversation history with every message, not by Claude actually remembering anything. Once you close that conversation window, the context vanishes entirely.

Claude's 200K token context window is impressive, but it is a buffer, not memory. It allows Claude to process large amounts of information in a single session, but nothing persists beyond that session. Project Knowledge in Claude.ai helps within a project scope, but it requires manual uploads, is limited in capacity, and does not work across Claude Desktop, Claude Code, or third-party integrations. You need something that works everywhere Claude works.

How PersistMemory Gives Claude Long-Term Memory

PersistMemory connects to Claude through the Model Context Protocol, the open standard that Anthropic designed specifically for extending Claude's capabilities. When connected, Claude gains three new abilities: it can store important information as memories, search through past memories using semantic similarity, and retrieve relevant context automatically when you ask questions. All of this happens transparently through MCP tool calls that Claude makes on its own.

The memory layer uses vector embeddings for semantic search, which means Claude does not need to match exact keywords. If you stored a memory about "the project uses PostgreSQL with pgvector," Claude can find it when you ask about "database configuration" because the system understands the semantic relationship between these concepts. This makes retrieval natural and reliable.

Setup: Claude Desktop Integration

Claude Desktop supports MCP natively. You configure servers through the claude_desktop_config.json file. On macOS this lives at ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows at %APPDATA%\Claude\claude_desktop_config.json, and on Linux at ~/.config/Claude/claude_desktop_config.json.

Add the PersistMemory server to your configuration:

{
  "mcpServers": {
    "persistmemory": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.persistmemory.com/mcp"
      ]
    }
  }
}

Save the file and restart Claude Desktop. On first launch, a browser window will open for you to log in with your PersistMemory account. This is a one-time OAuth authentication — your session lasts 20 days. The PersistMemory tools will appear under the hammer icon in your chat input area.

Setup: Claude Code Integration

Claude Code, Anthropic's terminal-based coding assistant, supports MCP servers directly from the command line. The setup is a single command:

# Add PersistMemory to Claude Code
claude mcp add persistmemory \
  --transport http \
  --url https://mcp.persistmemory.com/mcp

# Verify the connection
claude mcp list

Once connected, Claude Code remembers your codebase architecture, your preferred patterns, past debugging sessions, and deployment procedures across every terminal session. This is particularly powerful for developers who use Claude Code as their primary coding companion because it eliminates the need to re-explain project context at the start of each session.

Features That Matter for Claude Users

What to Store in Claude's Memory

Focus on information that provides lasting value across sessions. Project architecture decisions and tech stack choices are ideal candidates. Coding conventions, API patterns, database schemas, and deployment configurations are all high-value memories. Personal preferences like communication style, preferred frameworks, and coding patterns also compound in value over time.

Avoid storing ephemeral data like temporary debug values or one-time questions. The goal is to build a curated knowledge base that makes every future conversation with Claude more productive, not a raw log of everything that was ever discussed.