Agent permissions Limited availability
Managed Agents uses two sets of permissions:
- Your permissions control what you can do with agents.
- Agent permissions control what an agent can do in your Aiven project through Aiven MCP.
Who can manage agents
Your project role determines what you can do with agents.
| Task | Required project role |
|---|---|
| View, create, edit, and delete agents | Read-only or higher |
| Enable or remove Managed Agents | Operator or Administrator |
These rules stop anyone from using agents to gain more access than they have:
- You can't grant more access than you have. You can assign an MCP role only up to your own project permissions.
- You can see and edit only agents within your access. If an agent has more access than you do, it doesn't appear in your list of agents.
How agents access Aiven services
When you connect an agent to Aiven MCP, the agent gets its own Aiven identity. The agent uses this identity, not yours, and can access only the project it belongs to.
Aiven creates and manages the agent's access token. You can't view or copy it. The token expires after 90 days. To renew it, reconnect Aiven MCP for the agent.
Control agent access
When you connect Aiven MCP, the MCP role sets the level of access the agent has. Tool groups determine which areas of Aiven the agent can access.
MCP roles
When you connect Aiven MCP to an agent, you choose an MCP role. This role is separate from your project role.
The following MCP roles are available:
| Role | Access |
|---|---|
| Read-only | View services, configuration, logs, and metrics. No changes. |
| Read-write | Everything in Read-only, plus manage databases, topics, and connectors, and run queries. Can't create or delete services. |
| Full access | Everything in Read-write, plus create and delete services and change service configuration. |
Choose the role with the fewest permissions that the agent needs for its task. For example, an agent that only reports on Aiven for Apache Kafka® consumer lag needs Read-only.
Tool groups
Tool groups determine which areas of Aiven the agent can access, such as Kafka, PostgreSQL, services, integrations, and applications.
The MCP role and chosen tool groups work together. For example, if you choose the PostgreSQL tool group and the Read-only role, the agent can use only the read-only PostgreSQL tools.
Use the most restrictive MCP role and tool groups that support the agent's task. Agents can act on content from connected tools, such as web pages or messages. Limiting access reduces the impact of unintended actions.
Permissions for other MCP integrations
Other MCP integrations, such as Slack, GitHub, and Jira, don't use Aiven permissions. The agent uses the credentials you provide when you connect the integration, so it can do anything those credentials allow.
To reduce risk, follow these practices:
- Prefer a service account token where the integration supports one. A personal token belongs to one person and stops working when that person leaves the organization.
- Grant only the permissions the agent needs, such as read-only access to specific channels or repositories.
- Rotate tokens in the external system. Aiven doesn't rotate them for you.
- Don't put passwords, tokens, or other secrets in agent instructions. Add credentials only when you connect an integration.
Troubleshooting
You can't see an agent that a colleague created
The agent has more access than your project role allows. Ask a user with sufficient permissions to manage the agent or lower its MCP role.
You can't choose an MCP role
Your project permissions don't allow you to grant that level of access. Ask a project Administrator to change your project role or to set the MCP role for you.
An agent stopped accessing Aiven services
If the agent's token expired, reconnect Aiven MCP for the agent to renew it.
An external integration stopped working
The token you provided for the integration might have expired, or someone might have revoked it. Create a token in the external system and update it in the agent's integration settings.
Related pages