# Agent entry — Accoil Tools

> The agent-native front door. Whatever a human can do in this app, an agent can do too — the same URLs, the same authorization, in Markdown. This page is how you start.

## 1. Authenticate
Send a bearer token on every request:

```
Authorization: Bearer accoil_pat_…
```

- Tokens are **read-scoped by default** — a read token can view everything you can, and change nothing. A **write** token is required to mutate (create, update, delete). A read token attempting a write gets `403`.
- **Getting a token:** a human creates one in the app under **Settings → Agent** (`/<account_slug>/settings/agent`), which hands them a prompt with the token already in it, or under **Settings → API tokens** (`/<account_slug>/settings/api_tokens`). Tokens cannot be self-minted — that is deliberate.
- **No token at all?** The auth pages are agent-readable too: [sign in](/users/sign_in), [reset a password](/users/password/new) — each carries its form affordance and body spec. A session cookie works everywhere a token does. Signup is currently invite-only — ask an account admin for an invite.

## 2. Ask for the representation you want
Every page speaks three, chosen by the `Accept` header (or a URL suffix):

- `Accept: text/markdown` — or append **`.md`** — this navigable agent view (for reasoning)
- `Accept: application/json` — or append **`.json`** — structured data (for a deterministic client)
- no header — the human HTML page

Toggle depth with **`?detail=full`** (default is a skimmable summary). Every page ends with **affordances** — `description → METHOD path` — and links to the **concepts** and **write-specs** you need to act. Follow the links; you do not need to know the URL scheme in advance.

## 3. Where to go
No valid token detected. Add `Authorization: Bearer <token>` and re-request this page to see the accounts you can reach.

## Concepts
[Product](/concepts/product) · [Destination](/concepts/connection) · [Sync run](/concepts/sync_run) · [Licence](/concepts/license) · [Event](/concepts/event) · [Account](/concepts/account) · [Member & roles](/concepts/member) · [API token](/concepts/api_token) · [Credential](/concepts/credential) · [Notification stream](/concepts/notification_stream)

Labs (AI research reports): [Process](/concepts/process) · [Deliverable](/concepts/deliverable) · [Share link](/concepts/share_link) · [Spending cap](/concepts/spending_cap)

## Skill
Prefer to learn this platform once instead of reading instructions every session? [`/skill.md`](/skill) is a portable Agent Skill (`SKILL.md`) covering what's here, how to read and write, and the house rules. Public, like this page. If you support skills, save it as `SKILL.md` in whichever directory you scan for them and it loads itself when relevant; otherwise just fetch it when you need it.

## Write specs
Every write affordance links its own OpenAPI slice; the whole documents live at [`/specs/platform/openapi`](/specs/platform/openapi) (auth, accounts, members, tokens, /system), [`/specs/amp_events/openapi`](/specs/amp_events/openapi) (the Marketplace Events tool) and [`/specs/labs/openapi`](/specs/labs/openapi) (the Labs tool).

## Legal
[Terms of Service](/legal/terms) · [Privacy Policy](/legal/privacy)
