MCP Server

Overview

SKMemory exposes its full API through an MCP (Model Context Protocol) server, allowing any MCP-compatible AI agent to store and retrieve memories.

Starting the Server

# stdio transport (for local agents)
skmemory serve --transport stdio

# SSE transport (for network agents)
skmemory serve --transport sse --port 8765

Available Tools

skmemory_snapshot

Store a new memory.

{
  "title": "Meeting notes",
  "content": "Discussed the new architecture...",
  "tags": "meeting,architecture",
  "emotions": "focused,excited",
  "intensity": 7
}

skmemory_search

Search memories by keyword.

{
  "query": "architecture meeting",
  "limit": 10
}

skmemory_ritual

Run the rehydration ritual — restore identity, soul, and strongest memories on session start.

skmemory_context

Load a token-efficient memory context for prompt injection. Respects token budgets.

skmemory_recall

Retrieve full content of a specific memory by ID.

Integration Examples

OpenClaw

mcpServers:
  skmemory:
    command: skmemory
    args: [serve, --transport, stdio]

Claude Desktop

{
  "mcpServers": {
    "skmemory": {
      "command": "skmemory",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}