Lifous

Connect an AI assistant

Lifous as an MCP server: step-by-step set-up for Claude.ai, Claude Code and ChatGPT, a key for Cursor, and what to do when a step fails.

Lifous is an MCP server (Model Context Protocol, Streamable HTTP) at https://www.lifous.com/api/mcp. Point an assistant at it and it can list and write your to-dos, log habits, read and write notes, add flashcards and see what you are focusing on — through the same operations as the API, under the same limits. Two ways in: sign in (Claude, ChatGPT and Claude Code ask you once, in a popup — no key to paste) or an API key in a header (Cursor and scripts). Every plan can; Free is held to a smaller quota, enough to try it. The tools are listed on MCP tools.

Claude.ai and the Claude desktop app

  1. Open claude.ai (or the desktop app) and go to Customize → Connectors. On older layouts it is under Settings.
  2. Press Add custom connector.
  3. Name it Lifous and enter the URL https://www.lifous.com/api/mcp.
  4. If the dialog asks how to authenticate, choose Sign in now. If it only shows Advanced settings, leave the client ID and secret empty.
  5. If it asks about the OAuth client, choose Use Claude’s published identity. Register automatically works too; the first is the recommended one.
  6. Press Add. A Lifous page opens in a popup. If you are signed out of Lifous, sign in — the page comes back on its own.
  7. The page reads Connect Claude to Lifous, with From claude.ai under it, and says what Claude wants to do. Press Allow.
  8. The popup closes and the connector shows as connected. Open it: 41 tools are listed, each with a title. In a chat, switch the connector on under the + menu and ask “List my to-dos”.
What you enter
Name:            Lifous
URL:             https://www.lifous.com/api/mcp
Authentication:  Sign in now
OAuth client:    Use Claude's published identity

Claude Code

  1. In a terminal, add the server:
    claude mcp add --transport http focuscrew https://www.lifous.com/api/mcp
  2. Start a session and type /mcp. Pick focuscrew, then Sign in.
  3. Your browser opens the same Lifous page. It adds a line saying the app runs on your own computer — that is expected for Claude Code, which listens on a local port for the answer. Press Allow.
  4. Back in the terminal, /mcp shows focuscrew as connected. Ask Claude Code to list your to-dos.

ChatGPT

  1. Open chatgpt.com and go to Settings → Apps & Connectors.
  2. Open Advanced settings and turn on Developer mode.
  3. Back on the connectors page press Create — or Add custom MCP server, if that is what it offers. OpenAI moves this entry around; it may also sit behind the + button on chatgpt.com/plugins.
  4. Name it Lifous, enter the URL https://www.lifous.com/api/mcp, and choose OAuth for authentication. Leave the client ID and secret empty.
  5. Acknowledge the notice about custom servers and create it. ChatGPT opens the same Lifous page, reading Connect ChatGPT to Lifous. Press Allow.
  6. Check the tool list, then in a chat with the connector enabled ask “List my to-dos”.
What you enter
Name:            Lifous
MCP server URL:  https://www.lifous.com/api/mcp
Authentication:  OAuth

Afterwards

Everything you connected is listed in Settings → Integrations → Connected apps — Claude via claude.ai, ChatGPT via chatgpt.com — with when it last called. Disconnect stops it at once. A sign-in lasts: the access token is renewed in the background for up to ninety days of use, then the app asks you once more. A brand-new Lifous account lands on the welcome questions first; press Connect in the app again afterwards.

With an API key instead

For Cursor, scripts, and any client that cannot open a browser: make a key in Settings → Integrations → API keys and send it as a header.

Claude Code
claude mcp add --transport http focuscrew https://www.lifous.com/api/mcp \
  --header "Authorization: Bearer fc_live_…"
Cursor — .cursor/mcp.json
{
  "mcpServers": {
    "focuscrew": {
      "url": "https://www.lifous.com/api/mcp",
      "headers": { "Authorization": "Bearer fc_live_…" }
    }
  }
}
Claude.ai — Add custom connector, with a key
Authentication:  No sign-in
Request headers: Authorization = Bearer fc_live_…   (Required)
Claude Desktop, older builds — claude_desktop_config.json (through the mcp-remote bridge)
{
  "mcpServers": {
    "focuscrew": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://www.lifous.com/api/mcp", "--header", "Authorization: Bearer fc_live_…"]
    }
  }
}

If something goes wrong

  • The assistant says a tool is missing that this page lists. Apps keep the tool list from when the connector was added. Refresh the Lifous connector in the app’s settings — or remove it and add it again — and it reads the current 41.
  • The assistant says it hit a rate limit. A Free account has 20 requests a minute and 200 a day; a long conversation can reach it. Wait the seconds it names, or move to Pro or Max for 120 a minute and 20,000 a day.
  • “We couldn’t read the app’s details at claude.ai” (or chatgpt.com). Lifous fetches the app’s own description to check it; the fetch failed. Press Connect in the app again — nothing is remembered from a failure.
  • “The app asked to be sent somewhere it didn’t register.” The app sent a return address its description does not list. Lifous stops rather than sending your sign-in there. Remove the connector and add it again.
  • Claude says “Couldn’t reach the MCP server”. A moment’s outage on either side; try once more. The server answers every request without a sign-in with a standard 401 that names where to sign in — that is what Claude reads.
  • You disconnected, and the app still shows a connection. Its next call is refused. Remove it on the app’s side too, or sign in again to make a new connection.

Good to know

  • Any client that speaks MCP over HTTP works, by sign-in (OAuth 2.1 with PKCE; the server publishes its details at /.well-known/oauth-authorization-server, accepts Client ID Metadata Documents and registers other clients dynamically) or with a key in a header.
  • Nothing is kept between calls on our side — every request carries the credential, so disconnecting or revoking ends the connection at once.
  • The server tells the assistant to read before it writes, to say what it changed, and to ask before deleting; every tool says whether it reads, writes or deletes.