# Configure a Resource

> Add a capability, connect its authentication, and verify that it is ready for Agents.

## Before you begin {#before-you-begin}

Decide what the Agent needs to accomplish, not which internal component should perform it. Choose the closest category: AI & Models, Apps & APIs, Data, Storage, or Compute. For an external service, create the required [Credential](/docs/credentials) first and note any Policy restrictions that should apply. You need permission to create Resources in the target namespace.

## Steps {#steps}

1. Give the Resource a short, descriptive namespace-local name, such as `github-support`, `orders-readonly`, or `research-browser`.
2. Select the capability category and supported integration template through your platform setup workflow or the Resources API. For model access, configure a connection Gateway first and then configure one or more logical Models on top of it; see [Gateways and Models](/docs/resources/gateways-and-models.md).
3. Enter only the settings needed to reach the service, such as a service URL, database name, model choice, region, or capacity limit.
4. Attach a Credential reference when the external service requires authentication. Secret material remains in Credentials and does not belong in Resource settings.
5. Review the available capabilities and their interruption behavior. Pay particular attention to actions that change external state or may require outcome review.
6. Attach access Policy and enable the Resource when the review is complete.

## Verify {#verify}

Open **Resources**, search for the name, and confirm the expected category, status, and capability count. Open the detail page and check **Settings**, **Available capabilities**, and **Connections & access**. The detail page should show a canonical CRN and must not expose secret material. If the Resource depends on a Credential, verify that Credential separately without copying its value.

## Next steps {#next-steps}

Continue with [Use Resources from Agents](/docs/resources/use.md). For a tenant, customer, or principal-specific target, configure a scoped binding as described in [Scoped bindings](/docs/credentials/scoped-bindings) instead of duplicating Agent deployments.