---
title: 'Personal agents'
label: 'Personal agents'
description: 'Connect a personal agent such as Muse through the Personal Agent Protocol, choose exactly what it can see and change, and disconnect it at any time.'
publishedAt: 2026-11-17
order: 14
status: published
faqs:
  - question: Which Cavuno plans can connect a personal agent?
    answer: "Every paid Cavuno plan, starting with Starter, and the Starter trial. Personal agents use the Cavuno API, which the Free plan does not include."
  - question: Does the agent get my password?
    answer: "No. You sign in to Cavuno yourself and approve access on Cavuno's own page. The agent receives a token that only does what you approved."
  - question: Can a personal agent do more than my role allows?
    answer: "No. Every request is checked against the access you approved and against your current role on the board. If your role changes, the agent's access changes with it."
  - question: How do I disconnect an agent?
    answer: "Open Settings → Connected agents and select Disconnect next to the agent. It loses access straight away."
  - question: Does disconnecting an agent affect my teammates?
    answer: "No. Each member connects their own agents, and Connected agents shows only yours."
---

A personal agent is an AI assistant that works for you across many services, such as [Muse](https://muse.ai). Cavuno supports the [Personal Agent Protocol](https://personalagentprotocol.org) 1.0, so a personal agent can find Cavuno, ask you for access, and then work on your job board through the Cavuno API and MCP server.

You choose exactly what the agent can see and change, and you can disconnect it at any time.

You need a paid Cavuno plan (Starter or higher, or the Starter trial).

## Connect a personal agent

1. In your agent, ask it to connect to Cavuno, or to do something on your job board.
2. The agent opens Cavuno's sign-in page. Sign in to Cavuno.
3. Read what the agent asked for, listed under its name, such as **Muse asks to**. Each item says what it allows, and is marked **Can't be undone** or **Personal data** when that applies. Untick anything you don't want the agent to have.
4. Under **Choose an account**, select the board. This approves the connection.

The page also shows the website the agent comes from. An agent chooses its own name, so check the website before you approve.

## What you can allow

An agent asks only for what it needs, and you can approve less. "Manage" includes "See", and "Manage and publish" includes both.

| Access | What it allows | Marked |
|---|---|---|
| See board analytics | Traffic, apply clicks and search analytics for the job board | |
| See jobs | Job listings and their stats, including drafts and expired jobs | |
| Manage jobs | Create, edit, pause, expire, duplicate and delete jobs | Can't be undone |
| Manage and publish jobs | Also publish jobs so job seekers see them | Can't be undone |
| See companies | The companies that post on the board | |
| Manage companies | Create, edit, block and delete companies | Can't be undone |
| See employer accounts | Employer accounts and their contact details | Personal data |
| Manage employer accounts | Approve company claims, invite employers by email, deactivate or delete employer accounts | Can't be undone, Personal data |
| See candidates and applications | Job seekers, their profiles and their applications, including CVs | Personal data |
| Manage candidates and applications | Edit, deactivate and delete candidates, and update applications | Can't be undone, Personal data |
| Read the board inbox | Messages from candidates and employers | Personal data |
| Reply in the board inbox | Reply to messages by email and add notes | Can't be undone, Personal data |
| See job alert subscribers | Subscribers and their alerts | Personal data |
| Manage job alerts | Import subscribers, resend confirmation emails and unsubscribe people | Can't be undone, Personal data |
| See blog posts | Blog posts, authors and tags, including drafts | |
| Manage blog posts | Write, edit and delete blog posts, authors and tags | Can't be undone |
| Manage and publish blog posts | Also publish blog posts | Can't be undone |
| See categories and skills | Job categories, skills, locations and other ways jobs are grouped | |
| Manage categories and skills | Create, edit, merge and delete categories, skills and locations | Can't be undone |
| See uploaded files | Images and files uploaded to the board | |
| Manage uploaded files | Upload and delete images and files | Can't be undone |
| See job imports | Job imports and job feeds and how they ran | |
| Manage job imports | Set up, run and delete job imports and feeds | Can't be undone |
| See board settings | The board's settings | |
| See revenue and transactions | Revenue, payments and transactions, with buyers' names and email addresses | Personal data |
| See pricing plans and orders | The plans employers buy and their orders, with buyers' names and email addresses | Personal data |
| Follow exports and imports | Long-running exports and imports and their results | Personal data |
| Everything your role can do | Everything your role allows on the board | Can't be undone, Personal data |

Some actions need **Everything your role can do**: billing and pricing changes, team members and roles, domains, board settings changes, integrations, webhooks, site pages and redirects, job backfill, marketing permissions, the audit log, and deleting an application or an inbox conversation. Approve it only for an agent you trust with your whole account. Otherwise, make those changes in the dashboard.

API keys are managed only in the dashboard.

## See and disconnect your agents

Open **Settings → Connected agents** to see each agent you have connected, the website it comes from, the access you approved, and when it was last used. **Last used** updates each time the agent renews its access, about once an hour while it's working.

Select **Disconnect** to remove an agent. It loses access straight away, and has to ask you again to reconnect.

Connected agents shows only the agents you connected. Your teammates manage their own.

## For developers building a personal agent

Cavuno implements the Company side of Personal Agent Protocol 1.0.

- **Discovery:** `https://cavuno.com/.well-known/poppy.json`. It lists the MCP server and every scope, with descriptions in `auth.scopes` and hints in `auth.scope_hints`. The issuer is named in `https://cavuno.com/.well-known/oauth-protected-resource`.
- **APIs:** `poppy.json` lists the MCP server at `https://mcp.cavuno.com/mcp`, whose tools search and call the whole Operator API. To call the Operator API directly, use its OpenAPI description at `https://api.cavuno.com/v1/openapi.json`, with a token from the same sign-in requested without `resource` (see Tokens).
- **Issuer:** `https://api.cavuno.com/v1/oauth`. Its metadata is at `https://api.cavuno.com/.well-known/oauth-authorization-server/v1/oauth`.
- **Sign-in:** Direct Sign-In, the authorization code flow with PKCE. Your `client_id` is the URL of your client metadata document, and you authenticate at the token and revocation endpoints with `private_key_jwt`, signed with `ES256` or `PS384` by a key from your `jwks_uri`.
- **Tokens:** access tokens last 1 hour. Refresh tokens last 30 days and rotate on each use. Leave `resource` out of the sign-in, and each token request can choose: leave it out for the Operator API, or send `resource=https://mcp.cavuno.com/mcp` for the MCP server. A sign-in that names a resource gets tokens only for that resource.
- **Scopes:** request the narrowest scopes in `auth.scopes`. An Operator API request outside the approved scopes returns `403` with a `WWW-Authenticate` challenge (`error="insufficient_scope"`) that names the scope that would allow it, or `full_access` when no narrower scope does. Over MCP, the same refusal comes back as a tool error. `GET /v1/me` needs `full_access`.

For the MCP connection details, see [Cavuno MCP](/docs/mcp).
