Connecting your agent

rememd speaks MCP, so anything that speaks MCP can use it. The setup wizard writes the config for four clients; this is what it writes, and how to do it by hand.

rememd never launches your agent

You start Claude Code, Cursor or Codex however you already do. rememd is a tool server they connect to, not a wrapper around them. Context arrives through tool calls rather than through a file injected at startup — which is why there is nothing to configure per project.

rememd connect famnly -f "Track usage"   # bind this terminal
claude                                   # start your agent as usual

An agent that is already running picks the change up on its next rememd tool call, so you can connect in another pane without restarting it.

What the wizard writes

On first launch the app asks which agents to target and writes each one’s config itself. Four clients, four locations, and two formats that are not interchangeable.

Client	Config file	Format
Claude Code	`~/.claude.json`	JSON
Cursor	`~/.cursor/mcp.json`	JSON
VS Code	`~/Library/Application Support/Code/User/mcp.json`	JSON
Codex	`~/.codex/config.toml`	TOML

By hand, for a JSON client

~/.claude.json

{
  "mcpServers": {
    "rememd": {
      "command": "/usr/local/bin/rememd",
      "args": ["mcp-server"]
    }
  }
}

By hand, for Codex

~/.codex/config.toml

[mcp_servers.rememd]
command = "/usr/local/bin/rememd"
args = ["mcp-server"]

Chat clients and phones

A self-hosted backend also speaks MCP over HTTP at /mcp, so it can be registered at claude.ai as a custom connector and reached from Claude Desktop and the Claude app. That path spawns nothing locally, which is what makes it work on a device that cannot run a binary at all.

You cannot add a connector from a phone — register it on the web and it becomes available there. See Host it yourself.

Checking it took

rememd doctor        # diagnose stale or conflicting setup
rememd where         # what this folder resolves to
rememd ls workspaces