> ## Documentation Index
> Fetch the complete documentation index at: https://vms.dev/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# How programs and agents get access

> API keys, MCP sign-in, and agent registration: what each gives access to, and why.

There are three ways for software to get access. Each one answers a different question: who gives the access, and to what.

Explain which access method fits my program.

Ask what my program or agent must do, and whether a person is present.
Recommend one access method from this page and explain why.

## API keys belong to a workspace

An owner or an admin makes an [API key](https://vms.dev/docs/workspace/api-keys.md) for one workspace. The key acts for the workspace, not for a person. It is for a script or a CI job that runs without a person. Its scopes limit it to the REST API.

## MCP sign-in acts as a person

An [MCP client](https://vms.dev/docs/install-mcp.md), such as Claude or Cursor, opens sign-in in the browser. The person signs in and allows the access on a consent screen. The client then acts as that person, in each workspace that the person is a member of, and only with the person's role.

This needs the person at the browser when the client connects, and a client that supports OAuth.

## Agent registration starts from the agent

With [agent registration](https://vms.dev/docs/api-reference/agent-registration.md), the agent starts. It can arrive with no account and no key. It gets the same kind of token as an MCP client, so the same tools, roles, and limits apply.

The person still decides. The agent shows a code, and the person enters it after sign-in. The code goes from the agent to the person, and not in the other direction. Then a person who did not sign in cannot connect an agent to an account, and an agent that saw a code cannot use it. When the agent gives an email, only that account can enter the code.

An agent provider that already knows the person can skip the code. It signs an ID-JAG that says "this agent acts for this verified email". The service accepts it only from providers on its trust list. If the email already has an account, the person confirms the link one time.

## Each credential can stop

| Credential | It stops when |
| - | - |
| API key | An owner or an admin revokes it, or it expires |
| MCP sign-in | The access token expires after 15 minutes, and the person removes the client |
| Agent registration | The person chooses **Disconnect**, the provider sends a revocation event, or the 7-day assertion expires |

An unclaimed anonymous agent stops after 24 hours. It has no workspaces before a claim, so it can read no data.

Read the [changelog.](https://vms.dev/docs/changelog.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.