Recall ☁️

Persistent memory MCP server for AI coding agents.

Store, search, and retrieve context across sessions. Built on SQLite + FTS5.

Works with Claude Code, Cursor, GitHub Copilot, Gemini CLI, and any MCP-compatible agent.

Why?

AI coding agents are stateless. Every new session starts from zero. You re-explain your architecture, re-discover patterns, re-learn the same conventions.

Recall gives your agent a memory that persists. Store decisions, patterns, bug fixes, architecture notes — anything worth remembering. Next session, search and retrieve instantly.

Install

npm install -g recall-mcp

Or use directly with npx:

npx recall-mcp

Configure

Claude Code

Add to your .claude/settings.json:

{
  "mcpServers": {
    "recall": {
      "command": "npx",
      "args": ["recall-mcp"]
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "recall": {
      "command": "npx",
      "args": ["recall-mcp"]
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "recall": {
      "command": "npx",
      "args": ["recall-mcp"]
    }
  }
}

Tools

recall_store

Store or update a memory. If a memory with the same key exists, it gets updated.

key: "project.auth.decision"
content: "Using JWT with RS256. Public key rotated weekly via /keys endpoint."
tags: ["decision", "architecture", "auth"]
scope: "my-project"

recall_get

Retrieve a specific memory by key.

key: "project.auth.decision"
scope: "my-project"  // optional

recall_search

Full-text search across all memories. Returns ranked results.

query: "authentication pattern"
limit: 10
scope: "my-project"  // optional

recall_list

List recent memories, optionally filtered by scope.

scope: "my-project"  // optional
limit: 20

recall_delete

Delete a memory by key.

key: "project.auth.decision"
scope: "my-project"  // optional

recall_stats

Show memory statistics — total count, breakdown by scope.

How It Works

• Storage: SQLite with WAL mode. All data stored locally in ~/.recall/recall.db.

• Search: FTS5 full-text search with automatic ranking. Fast even with thousands of memories.

• Scoping: Memories are scoped to global by default. Use project-specific scopes to keep project context separate.

• Auto-indexing: Every store/update/delete automatically syncs the FTS index via SQLite triggers.

Use Cases

• Architecture decisions — "Why did we choose X over Y?" → stored and searchable

• Bug fix patterns — "This weird null pointer was caused by..." → never debug the same thing twice

• Code conventions — "In this project, we use..." → agents follow your rules automatically

• Project context — Store project structure, key files, gotchas for instant onboarding

• Cross-session learning — Agent remembers what it learned last session

Configuration

License

MIT