GemFit.ai MCP setup

GemFit provides a secure Model Context Protocol server for approved AI
assistants. It lets users connect Claude Code, OpenAI/Codex, OpenClaw, LM
Studio, MCP Inspector, and other compatible clients to their own GemFit fitness
account.

Remote MCP endpoint:
https://gemfit.ai/api/gemfit/mcp

OAuth metadata:
https://gemfit.ai/.well-known/oauth-protected-resource/api/gemfit/mcp
https://gemfit.ai/.well-known/oauth-protected-resource
https://gemfit.ai/.well-known/oauth-authorization-server

Human docs:
https://gemfit.ai/mcp
https://gemfit.ai/developers
https://gemfit.ai/docs

Claude Code setup:
claude mcp add --transport http gemfit https://gemfit.ai/api/gemfit/mcp

Claude app + mobile setup:
1. On Claude web, open Customize > Connectors > Add custom connector.
2. Choose HTTP transport and paste https://gemfit.ai/api/gemfit/mcp.
3. Complete GemFit OAuth once in the browser.
4. Open Claude mobile and enable GemFit from the chat connector/tools menu.

Claude mobile can use the account-level connector after web setup. Add new
custom connectors from Claude web.

Generic HTTP MCP client config:
{
  "name": "GemFit MCP",
  "transport": "http",
  "url": "https://gemfit.ai/api/gemfit/mcp"
}

LM Studio mcp.json:
{
  "mcpServers": {
    "gemfit": {
      "url": "https://gemfit.ai/api/gemfit/mcp"
    }
  }
}

Expected auth behavior:
The unauthenticated MCP endpoint returns a 401 Bearer challenge with
resource_metadata. The user approves OAuth scopes in GemFit Connect. A 401
before OAuth means authentication is required, not that the server is missing.
No personal API key is needed for standard clients. OAuth creates scoped,
revocable access for the signed-in user.

Important first calls:
1. gemfit_get_profile_context
2. gemfit_get_day_summary

Read tools:
- gemfit_get_profile_context
- gemfit_get_day_summary
- gemfit_search_logs
- gemfit_get_goals
- gemfit_get_recent_trends
- gemfit_get_data_quality_report

Draft tools:
- gemfit_create_food_log_draft
- gemfit_create_workout_log_draft
- gemfit_build_workout_plan_draft
- gemfit_create_body_metric_draft
- gemfit_create_trackable_log_draft
- gemfit_create_fitness_journal_draft
- gemfit_create_goal_update_draft
- gemfit_update_log_draft
- gemfit_delete_log_draft
- gemfit_list_pending_actions
- gemfit_confirm_pending_action
- gemfit_reject_pending_actions

Use pattern:
If the user asks to log meals or workouts, read profile context first, read the
day summary, audit data quality when cleaning existing data, then create review
drafts. Use profile/body metrics for workout calorie burn. Same-day workouts
belong in workout log drafts, not future plan drafts. Return pending review
links. Do not save durable health data until the user approves in GemFit Connect
or explicitly approves a specific pending draft in chat with logs.confirm. If the
user rejects stale drafts, use the reject tool instead of leaving the review
queue messy.

Security:
GemFit MCP exposes only the authenticated user's fitness data. It has no business/admin, billing admin, deployment, schema, service-role, env var, role-management, email, or other-user tools. Tools never accept caller-controlled user_id. Every call is audited.

Client guidance:
Public GemFit pages are documentation only. The MCP server, OAuth scopes, and
tool schemas are authoritative for connected-client behavior.