Connect your agents

export COMMONPLACE_URL=https://your-server/mcp
export COMMONPLACE_TOKEN=<from your setup page>

Claude Code

The recall skill fires on its own when a turn needs memory. Stop and PreCompact hooks file the session.

claude mcp add --transport http commonplace "$COMMONPLACE_URL" \
  --header "Authorization: Bearer $COMMONPLACE_TOKEN"

Codex

The recall skill fires on its own when a turn needs memory. Stop and PreCompact hooks file the session.

# ~/.codex/config.toml
[mcp_servers.commonplace]
url = "https://your-server/mcp"
bearer_token_env_var = "COMMONPLACE_TOKEN"

Cursor

The sessionStart hook tells the agent to search the workspace's wing before it answers. The recall skill and an optional rule cover the turns after it.

// ~/.cursor/mcp.json
{"mcpServers": {"commonplace": {
  "url": "https://your-server/mcp",
  "headers": {"Authorization": "Bearer ${env:COMMONPLACE_TOKEN}"}
}}}

Antigravity

On the first model call the wake hook injects the palace's own words, verbatim. The recall skill and rule take over from there.

// ~/.gemini/antigravity/mcp_config.json
{"mcpServers": {"commonplace": {
  "serverUrl": "https://your-server/mcp",
  "headers": {"Authorization": "Bearer <token>"}
}}}

Cloud clients

claude.ai, ChatGPT and Grok connect over OAuth, so your server needs a public HTTPS address. The consent page on your server picks which wings each one may read. Their writes land in the inbox.

claude.ai   Settings → Connectors → Add custom connector → paste the URL
ChatGPT     Settings → Connectors → developer mode → create → paste the URL
Grok Bot    add a connector by URL, one connection per bot

https://your-server/mcp

Local models

LM Studio, or Open WebUI in front of Ollama, over your LAN or tailnet. The model searches when you ask it to.

// LM Studio mcp.json
{"mcpServers": {"commonplace": {
  "url": "http://your-server:8765/mcp",
  "headers": {"Authorization": "Bearer <token>"}
}}}