Connecting your client
Wadachi is a standard MCP server over stdio. The whole contract: run the command
wadachi (no arguments) with the environment variable BRAIN_DIR pointing at your
brain. Any MCP-compatible client works.
Claude Code
wadachi init registers it for you. Manually:
claude mcp add wadachi -e BRAIN_DIR=$HOME/.wadachi -- wadachi
Check with claude mcp list — you should see wadachi ✔ Connected.
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"wadachi": {
"command": "wadachi",
"args": [],
"env": { "BRAIN_DIR": "/Users/you/.wadachi" }
}
}
}
Cursor
.cursor/mcp_servers.json — same JSON shape as Claude Desktop.
Antigravity
~/.gemini/antigravity-ide/mcp_config.json — written by wadachi init; same shape.
Any other MCP client
Point it at the wadachi executable with BRAIN_DIR in the environment. If your
client can't set env vars, wadachi falls back to ~/.wadachi (or a legacy
~/.engram if present).
One brain, many machines
BRAIN_DIR is the only pointer, so a brain on an external drive or synced folder can
serve several machines. The safe pattern is one writer at a time — SQLite handles
locking, but two agents writing simultaneously from different machines over network
filesystems is asking for trouble.
Teaching your agent to use it
The server ships instructions to the model automatically (call get_context first,
store as you go, link memories). If you use a CLAUDE.md, a line like "start every
session with wadachi's get_context" makes the habit explicit. See get_context — start here, every session.