# Agents

Claude, ChatGPT, Cursor or an agent of your own can work in an Acrune
workspace as a person on its team, through the hosted MCP server at
`https://track.acrune.com/mcp` (streamable HTTP).

An agent is never more than the person it acts for. Every tool is bound to
that person's workspace and re-checks their seat on every call; a token
whose person leaves the workspace stops working. Tools that change
anything need a token allowed to write, and only owners and admins can
have one.

## Connecting

**With a sign-in (OAuth).** Add `https://track.acrune.com/mcp` as a custom
connector in Claude.ai, ChatGPT or any client that speaks the MCP
authorization spec. The client registers itself, sends you to
`acrune.com/dashboard/oauth/authorize` to choose a workspace and approve,
and receives a token that lasts 30 days and renews itself.

**With a token.** In the dashboard, Settings > Agents makes a personal
token (`acr_agent_…`), shown once:

```bash
claude mcp add --transport http acrune https://track.acrune.com/mcp \
  --header "Authorization: Bearer acr_agent_…"
```

Other clients take the same address and header.

## Tools

| Tool | What it does | Writes |
| --- | --- | --- |
| `whoami` | The person, workspace and role the token acts as |  |
| `list_campaigns` | Campaigns, newest first, with their creator links |  |
| `campaign_stats` | Clicks, creators, conversions and value, submissions and earnings, over a period |  |
| `search_creators` | Creators by name or handle |  |
| `list_applications` | Discord onboarding applications |  |
| `decide_application` | Approve, reject or ban one, as the review card's buttons do | yes |
| `list_form_applications` | Applications made through the application form |  |
| `list_verifications` | Audience verifications and what their recordings showed |  |
| `decide_verification` | Approve or reject one, granting the demographics stage | yes |
| `list_tickets` | Discord tickets from the workspace's ticket panels |  |
| `reply_ticket` | Post in a ticket as the server's bot, signed as an agent | yes |
| `list_payouts` | Payouts to the workspace's creators, and any that failed |  |

Decisions an agent makes are recorded as `<person> (agent)`, so the team
can always tell them apart from their own.

## Discovery

- Protected resource metadata:
`https://track.acrune.com/.well-known/oauth-protected-resource/mcp`
- Authorization server metadata:
`https://track.acrune.com/.well-known/oauth-authorization-server`
- The HTTP API, for integrations rather than agents:
`https://track.acrune.com/openapi.json`
- This documentation for language models: `/docs/llms.txt`, or all of it
in one file at `/docs/llms-full.txt`.
