Connect Freddy from any AI agent

Freddy now supports the standard device-flow sign-in (RFC 8628), the same flow GitHub Copilot CLI, AWS SSO, and smart TVs use. Any agent that speaks the standard can sign you in, connect wearables, query your data, and manage your subscription, with no copy-pasted URLs and no manual config files.

This article is the umbrella. For a specific agent, see Connect Freddy from Claude Code or Connect Freddy from ChatGPT. If something does not work, see agent troubleshooting.

For Claude Desktop or claude.ai (the consumer apps), use the simpler Connect Freddy to Claude walkthrough instead.

How the new sign-in works

  1. The agent registers itself with Freddy (one HTTP call).
  2. The agent asks Freddy for a short-lived sign-in URL.
  3. The agent shows you that URL. You open it in any browser, approve, done.
  4. The agent gets a sign-in token and starts working.

You sign in once. The agent can then do anything you can do in the dashboard.

What works today

As of May 2026, agents fall into three buckets:

  1. ChatGPT custom connectors sign you in automatically through the existing redirect flow. No extra steps needed. See Connect Freddy from ChatGPT.
  2. Claude Code, Cursor, your own bot, and other CLI or server-side agents can use device flow, but most of them do not yet know about Freddy specifically. You will either need to (a) tell the agent about the flow in its system instructions, or (b) run a 5-line shell recipe once and paste the resulting access token into the agent's MCP config. The Claude Code article shows both paths. The shell recipe works the same way for Cursor and any other agent that lets you set a custom Authorization header on an MCP server. We do not have a dedicated Cursor walkthrough yet.
  3. Claude Desktop and claude.ai keep working as before through the existing connector UI.

We expect agent vendors to ship native device-flow support over time. When they do, this gets even simpler. Until then, the recipes below give you a working setup today.

The shell recipe (works for any agent)

If your agent does not walk the device flow itself, you can run this once in a terminal and paste the resulting access token wherever your agent expects an MCP bearer token. The token is good for 1 hour, and the recipe also returns a 60-day refresh token for renewal.

APP=https://freddy.coach
CLIENT=$(curl -sX POST $APP/oauth/register \
  -H "Content-Type: application/json" \
  -d '{"client_name":"My Agent","redirect_uris":["http://localhost"]}' \
  | jq -r .client_id)

DA=$(curl -sX POST $APP/oauth/device_authorization \
  --data-urlencode "client_id=$CLIENT" \
  --data-urlencode "scope=mcp account:read connections:write")

DEVICE=$(echo "$DA" | jq -r .device_code)
URL=$(echo "$DA"   | jq -r .verification_uri_complete)
echo "Open: $URL"

while :; do
  R=$(curl -sX POST $APP/oauth/token \
    --data-urlencode "grant_type=urn:ietf:params:oauth:grant-type:device_code" \
    --data-urlencode "client_id=$CLIENT" \
    --data-urlencode "device_code=$DEVICE")
  ERR=$(echo "$R" | jq -r .error)
  case "$ERR" in
    authorization_pending|slow_down) sleep 6 ;;
    null) echo "$R" | jq -r .access_token; break ;;
    *) echo "stop: $ERR"; break ;;
  esac
done

The recipe prints a sign-in URL, you open it in a browser, approve, and the script prints freddy_at_.... That string is the access token.

What the agent can do once signed in

Anything you can do in the dashboard:

The agent uses Freddy's MCP server (for queries and the three write tools connect_wearable, disconnect_wearable, sync_provider) plus a small REST surface for the rest. Developers building an agent integration should see docs/agent-quickstart.md in the repo.

Privacy and revocation

Each sign-in produces a grant tied to your Freddy account. You can review and revoke connected agents from the dashboard at freddy.coach/app/connections under "Connected apps". Revoking the grant immediately invalidates the agent's tokens.

Need help?

Email support@freddy.coach, or message us in-app at freddy.coach/app/chat.