# Tools and integrations

Source: https://www.hyperagent.com/docs/concepts/tools-and-integrations

How an agent takes action: built-in tools, connected apps, and skills for APIs that need a method.

An agent becomes useful when it can **take action**: search the web, query a warehouse, read a file, or post a result to Slack. In Hyperagent those actions come from built-in **tools** and connected **integrations** you allow on the agent.



## What they are [#what-they-are]

A built-in Hyperagent capability such as search, browser, code, images, webpages, tables, or maps. Toggle it on the agent.

A connected external account the agent can act through, such as Slack, Gmail, GitHub, Sheets, a CRM, a warehouse, or a custom MCP server.

## Pick the right path [#pick-the-right-path]

Enable a **tool** on the agent. The core capability does not need third-party account setup.

Connect an **integration** under Settings, then allow it on this agent. App setup lives under [Integrations](/docs/integrations).

Attach a remote MCP server your company or vendor operates. [Custom MCP setup](/docs/integrations/custom-mcp).

Create a [skill with scripts and credentials](/docs/concepts/skills/skills-edit-and-import) so the agent knows both how to call the API and how your team expects it to be used.

## Allow access on the agent [#allow-access-on-the-agent]

### Toggle built-in tools [#toggle-built-in-tools]

Enable only the capabilities the role needs. A research agent may need Search and Browser; a reporting agent may also need Tables and Documents.

### Connect external accounts [#connect-external-accounts]

Connect the integration at the account level. This authorizes Hyperagent to reach the service.

### Allow the integration on this agent [#allow-the-integration-on-this-agent]

Decide which connected apps this role may use, then narrow channels or resources where the integration supports scopes.

### Attach custom MCP only when needed [#attach-custom-mcp-only-when-needed]

Use this for a remote internal or vendor MCP server that is not already listed as a native integration.

A schedule that should post to Slack can research correctly and still fail to deliver if Slack is connected to your account but not allowed on that agent.

## Custom MCP integrations [#custom-mcp-integrations]

Hyperagent connects to a remote MCP server and calls its tools. Use an HTTPS
endpoint and the authentication method the product asks for. [Configure
Custom MCP](/docs/integrations/custom-mcp).

**Supported:** remote HTTPS MCP servers, including OAuth-backed servers when configured through the product.

**Not supported as a paste-anything path:** local stdio-only servers, arbitrary local commands, or endpoints that cannot be reached securely over HTTPS.

If a service offers neither a native integration nor a remote MCP server, use a skill with API credentials.

## Unattended runs [#unattended-runs]

Schedules and [Live Mode](/docs/concepts/agents/invocations/live-mode) can restrict external writes until you allow them. If an unattended run read the right data but never changed the connected app, inspect the agent's integration access, resource scope, unattended-write setting, and the run's tool trail.

## FAQs [#faqs]

Slack is an integration. Web search is a tool. Slack can also be an [invocation](/docs/concepts/agents/invocations) when a channel message starts the run.

Use a skill when the agent needs a repeatable method, a required output format, or an API path with scripts and credentials. A skill can teach an agent how your team uses Google Docs without turning Google Docs into a different kind of integration.

Use Custom MCP when your company or a vendor provides a remote MCP server
that is not already available as a native integration. Hyperagent connects
to that server and calls its tools.

Check [Integrations](/docs/integrations). If the service exposes a remote MCP server, add it as Custom MCP. If it only exposes an API, create a skill with scripts and credentials.

Usually no. Search, browser, code, tables, documents, and media generation are Hyperagent tools. A tool still must be enabled on the agent before it can be used.