CLI + headless MCP

Build agents on
real health data.

freddy is a personal MCP server for health and performance data. Sign a user up, connect their data sources, and query sleep, HRV, recovery, and workouts. All over the MCP standard, with an OAuth device flow built for headless clients and a one-command CLI.

npx @freddy-coach/cli login
OAuth device flow5 scopesread + write tools
Why freddy for agents

One server. Any agent.

Real health data over an open standard, with the auth and write tools a headless client needs.

Real data, 28+ sources

Polar, Oura, Garmin, WHOOP, Withings and more. The user's own accounts, never simulated.

The MCP standard

No proprietary API to learn. Any client that speaks MCP can read a user's data with one bearer token.

Device flow for headless

RFC 8628 device authorization plus dynamic client registration. No redirect_uri, no browser callback to wrangle.

Scoped, auditable write tools

Connect, disconnect, and sync data sources conversationally. Every call is scoped and lands in the user's audit log.

What agents do with it

Health context turns agents
into coaches.

Three patterns people actually run. Each one is a scheduled agent, a freddy query, and a message back to you.

every morning · 07:00

Morning briefing

A cron fires, the agent reads your night, and a message is waiting before your coffee.

agent · to you

HRV is 12% under your 7-day average and you only got 74 minutes of deep sleep. Take today easy; tomorrow looks good for intervals.

watches weekly · alerts on drift

Trend watchdog

The agent tracks the slow signals you forget to check, and only speaks when something moves.

agent · to you

Your resting heart rate has crept up 4 bpm over two weeks while sleep stayed stable. Worth a look at training load.

sundays · 18:00

Weekly report

A coach-style summary of the week across every source you have connected.

agent · to you

Training load up 18% on last week, sleep debt about an hour, HRV trending flat. One hard session fits before Thursday.

Recipes for your agent

Give it something
worth running.

Recipes an agent can run on a schedule, from the public library. Wire one into a cron and let it report back.

The full recipe library works for agents too.Browse all recipes →
Quickstart

Three commands.

The CLI wraps device-flow auth, token storage and refresh, the data-source connect handoff, and MCP calls.

npx @freddy-coach/cli login
npx @freddy-coach/cli connect oura
npx @freddy-coach/cli query --metrics hrv,sleep_duration --days 7

No terminal? Any MCP client can drive freddy over raw HTTP: register a client, start the device flow, poll for a token, then call /mcp. See the headless setup guide.

Scopes

Ask for exactly what you need.

The user approves scopes at sign-in. Omit scope and you get mcp only.

ScopeGrants
mcpMCP tool access: the read tools plus the three write tools.
account:readProfile, plan and subscription, connections, OAuth grants, email prefs, sync status.
account:writeRotate MCP token, email prefs, billing checkout/cancel, integration requests, chat.
connections:writeConnect, disconnect, and sync data sources.
audit:readAudit log plus bulk health-data export.
MCP tools

Read and write, conversationally.

All under the mcp scope. Queries plus three write tools for managing data sources.

ToolPurpose
get_profileUser metadata, connected data sources, plan.
list_metricsAvailable metrics with counts, date ranges, units, and sources.
query_metricsFetch metrics by name and date range.
connect_sourceReturns a verification URL for OAuth providers, or hands off to the freddy dashboard for API-key sources.
disconnect_sourceDisconnects and wipes synced data. Destructive.
sync_sourceTriggers an incremental sync. Fire-and-forget.
accept_termsRecords the user's acceptance of the current Terms and Privacy version.
Go deeper

Everything you need to ship.