@db0-ai/claude-code

Why

Without memory, every Claude Code session starts from zero. You explain your architecture, your preferences, the bug you debugged yesterday — and tomorrow you'll explain it all again.

With db0, a 10-minute context-building ritual becomes instant. Claude already knows your project uses PostgreSQL, that you prefer functional style, and that the CORS fix from last week was the Access-Control-Allow-Credentials header. No signup, no API keys, no cloud — your data stays in a local SQLite file.

Quick Start

claude mcp add --transport stdio db0 -- npx -y @db0-ai/claude-code

That's it. Restart Claude Code. No account, no API key, no config file.

Or add to .mcp.json:

{
  "mcpServers": {
    "db0": {
      "command": "npx",
      "args": ["-y", "@db0-ai/claude-code"]
    }
  }
}

How It Works

Just talk naturally. Claude uses the memory tools automatically:

You: "I always use bun, not npm. And remember that this project uses tRPC."
→ Claude stores both as user-scoped facts — visible in every future session.

You: "We switched from Heroku to AWS last week."
→ Claude supersedes the old Heroku fact. The old one is preserved for audit
  but excluded from search. No contradictions.

Next day, new session:
You: "Help me deploy this service"
→ Claude already knows: bun, tRPC, AWS. No re-explanation needed.

You don't need to say "call db0_memory_write." Just state your preferences, make decisions, or say "remember that..." — Claude uses the tools when it recognizes something worth remembering.

What You Get

• Zero setup, zero cost — local SQLite, no API key, no cloud account, no usage limits
• Your data stays on your machine — nothing leaves localhost unless you choose Postgres
• Claude remembers across sessions — preferences, decisions, project context persist
• Facts stay current — when things change, old facts are superseded with full audit trail
• Semantic search — Claude finds relevant memories by meaning, not just keywords
• Memory consolidation — related facts are merged over time, keeping memory clean
• Full visibility — inspector UI and CLI to browse what Claude knows

Tools

| Tool | Description | |---|---| | db0_memory_write | Store a fact with scope, tags, and optional superseding | | db0_memory_search | Semantic search with scope/tag filtering | | db0_memory_update | Update a fact — finds the old version, supersedes it, writes the new one | | db0_memory_list | List memories by scope | | db0_memory_get | Get a specific memory by ID | | db0_memory_delete | Delete a memory | | db0_memory_stats | Memory statistics by scope and status | | db0_memory_consolidate | Run memory maintenance — merge duplicates, promote chunks, clean edges | | db0_state_checkpoint | Create a state checkpoint | | db0_state_restore | Restore the most recent checkpoint | | db0_log_query | Query structured log entries |

Skills

/db0:inspect dark mode preference     # browse what Claude remembers
/db0:ingest I prefer TypeScript       # tell Claude to remember something

Memory Scopes

| Scope | Lifetime | Use case | |---|---|---| | user | Permanent, cross-session | Preferences, decisions, personal context | | agent | Permanent, all sessions | Agent-specific patterns, learned behaviors | | session | Current session | In-progress decisions, temporary context | | task | Current task | Scratch work, intermediate results |

Configuration

| Variable | Default | Description | |---|---|---| | DB0_STORAGE | ~/.claude/db0.sqlite | Storage path or PostgreSQL connection string | | DB0_AGENT_ID | claude-code | Agent identifier | | DB0_USER_ID | OS username | User identity for scope isolation |

Cross-device memory via PostgreSQL

{
  "mcpServers": {
    "db0": {
      "command": "npx",
      "args": ["-y", "@db0-ai/claude-code"],
      "env": {
        "DB0_STORAGE": "postgresql://user:pass@your-host/db0"
      }
    }
  }
}

Inspector

npx @db0-ai/inspector --db ~/.claude/db0.sqlite

Browse, search, and manage Claude's memories in a web UI. See @db0-ai/inspector.

vs. Mem0 for Claude Code

| | db0 | Mem0 | |---|---|---| | Setup | npx — one command, done | pip install + API key + cloud signup | | Cost | Free, unlimited | Free tier: 10K memories, 1K retrievals/month | | Data | Local SQLite on your machine | Cloud (app.mem0.ai) | | Fact correction | Superseding with audit trail | Overwrite (history lost) | | Extraction | Zero LLM calls (rules-based) | LLM call on every write | | State management | Checkpoints + branching | None | | Ecosystem | Same DB works with AI SDK, LangChain, OpenClaw, inspector | Standalone API | | Graph memory | Typed edges (free) | Pro plan required |

Both solve the same core problem. db0 is local-first and free. Mem0 is cloud-first with a managed API.

Manage

npx @db0-ai/openclaw upgrade claude-code      # upgrade
npx @db0-ai/openclaw uninstall claude-code     # uninstall

Documentation

• Comparison with other Claude Code memory tools

Part of db0

This MCP server is one entry point to the db0 SDK. The same memory database works with the core SDK, OpenClaw plugin, CLI, and inspector.

License

MIT