claw-recall（爪回忆）

Claw Recall — Searchable Conversation Memory for AI Agents

Your agent just lost context mid-task. The decision you made an hour ago? Gone. What your other agent figured out yesterday? Unreachable. Claw Recall fixes this by indexing every conversation into a searchable database your agents can query anytime.

The Problem

Context compaction drops critical decisions. Cross-session knowledge vanishes. Long conversations push early context out of the window. If you run multiple agents, they can't access each other's conversations at all. MEMORY.md helps with preferences, but it can't answer "what exactly did we discuss about the API last Tuesday?"

What Claw Recall Does

• Post-compaction recovery: Get the full transcript from before compaction wiped your context
• Cross-agent search: Any agent can search any other agent's conversations
• Unified search: Conversations, captured thoughts, Gmail, Google Drive, and Slack in one query
• Hybrid search: Keyword (FTS5) + semantic (OpenAI embeddings) with automatic detection
• Self-hosted: Your data stays on your machine. No cloud, no subscription, no vendor lock-in.

Installation

Claw Recall is an MCP server. Install the Python package, then connect it to your agent.

git clone https://github.com/rodbland2021/claw-recall.git
cd claw-recall
pip install -r requirements.txt
python3 -m claw_recall.indexing.indexer --source ~/.openclaw/agents-archive/ --incremental

Full setup guide: https://github.com/rodbland2021/claw-recall#quick-start

Connect via MCP (OpenClaw)

Add to your OpenClaw config (~/.openclaw/openclaw.json or agent config):

{
  "mcpServers": {
    "claw-recall": {
      "command": "python3",
      "args": ["-m", "claw_recall.api.mcp_stdio"],
      "env": { "PYTHONPATH": "/path/to/claw-recall" }
    }
  }
}

Connect via MCP (Claude Code)

Add to ~/.claude.json:

{
  "mcpServers": {
    "claw-recall": {
      "command": "python3",
      "args": ["-m", "claw_recall.api.mcp_stdio"],
      "env": { "PYTHONPATH": "/path/to/claw-recall" }
    }
  }
}

Remote agents (SSE)

Start the SSE server on the Claw Recall machine, then connect from anywhere:

python3 -m claw_recall.api.mcp_sse

claude mcp add --transport sse -s user claw-recall "http://your-server:8766/sse"

MCP Tools Reference

Primary Tools (use these most)

search_memory — The main search tool. Searches ALL sources in one call: conversations, captured thoughts (Gmail, Drive, Slack), and markdown files.

search_memory query="what did we decide about the API" [agent=butler] [days=7]

Optional params: agent (filter by agent name), days (limit to recent), force_semantic (use embeddings), force_keyword (use FTS5 only), convos_only, files_only, limit.

browse_recent — Full transcript of the last N minutes. The go-to tool for context recovery after compaction.

browse_recent [agent=kit] [minutes=30]

Returns the complete conversation with timestamps. Use this FIRST after any context reset.

capture_thought — Save an insight, decision, or finding so any agent can find it later.

capture_thought content="SQLite WAL mode requires checkpoint for readers to see writes" [agent=kit]

Secondary Tools

When to Use Each Tool

How It Works

Session Files (.jsonl) → Indexer → SQLite DB (FTS5 + vectors) → MCP Tools
                                        ↑
Gmail / Drive / Slack → Source Poller ───┘

All data is stored locally in a SQLite database. Keyword search uses FTS5 (zero API keys). Semantic search uses OpenAI embeddings (requires OPENAI_API_KEY in .env).

Requirements

• Python 3.10+
• SQLite 3.35+ (bundled with Python)
• OpenAI API key (optional, only for semantic search)

Links

• GitHub: https://github.com/rodbland2021/claw-recall
• Discord: https://discord.gg/4wGTVa9Bt6
• Full Guide: https://github.com/rodbland2021/claw-recall/blob/master/docs/guide.md
• Changelog: https://github.com/rodbland2021/claw-recall/blob/master/CHANGELOG.md