MCP server for the SoulSketch Protocol: gives any MCP-capable AI client (Claude Desktop, Windsurf, Cursor, …) tools to read, verify, and append to portable, user-owned AI memory packs.

The six tools

Tool	In plain English
`soulsketch_read_pack`	"Wake up and remember who you are" — load an identity
`soulsketch_observe`	"Write that down" — append one diary line (the only write)
`soulsketch_validate_pack`	Health-check a 5-file memory pack
`soulsketch_fingerprint_pack`	Tamper-evident SHA-256 seal of the pack state
`soulsketch_diff_packs`	Which identity dimension changed between two packs
`soulsketch_continuity_record`	Notarized snapshot: fingerprint + provenance labels

Setup

// in your MCP client config
{
  "mcpServers": {
    "soulsketch": {
      "command": "npx",
      "args": ["-y", "@soulsketch/mcp-server"],
      "env": {
        "SOULSKETCH_ALLOWED_ROOTS": "/path/to/your/soul-sanctum"
      }
    }
  }
}

A Docker image and a plain-language guide (including what each tool affords you and the full trust model) live in the protocol repository.

Trust model, in one paragraph

The server only touches directories you allow via SOULSKETCH_ALLOWED_ROOTS - normally your Soul-Sanctum, the private repo where your assistant's memory packs live. Five of the six tools are read-only; the sixth appends single lines to runtime_observations.jsonl and can never modify or delete existing memories. No network calls, no telemetry, no hosted anything: your assistant's memory is a folder of text files on your computer, in your Git history.