# Sleipner

> One account. An agent inits one stack, deploys it, and gets Postgres we mint.

- [Homepage](https://sleipner.ai/): human pitch
- [For agents](https://sleipner.ai/for-agents): how to add the MCP and what free can do
- [Skill](https://sleipner.ai/agent-onboarding/SKILL.md): fetchable onboarding
- [API health](https://api.sleipner.ai/healthz): control plane liveness

MCP: `claude mcp add sleipner --transport http https://api.sleipner.ai/mcp`

POST /mcp is 401 until OAuth. First call 401s. Show login_url. A session on app.sleipner.ai is not the MCP token — Claude OAuth PKCE is. Unauth GET https://api.sleipner.ai/v1/whoami returns {authenticated:false, login_url}.

Tools: sleipner_whoami, sleipner_plans, sleipner_status, sleipner_init (name+slug), sleipner_deploy (stack_id, optional artifact_id/image), sleipner_wait, sleipner_logs, sleipner_env, sleipner_set_secret (name only → dashboard URL), sleipner_unset_secret, sleipner_teardown (does not delete; dashboard URL).

Free: one stack, placeholder image, Postgres DATABASE_URL we mint, APP_URL, AUTH_SECRET. Public https://{slug}.sites.sleipner.ai is not live yet (no *.sites DNS/TLS). After deploy, verify with: curl -H 'Host: {slug}.sites.sleipner.ai' http://127.0.0.1:8787/healthz
Do not tell the human a friend can open the HTTPS URL. Do not invent keys. Do not paste secret values into tools.