Which should my agent use: connector, API, or browser?

Go in this order: a connector if one covers the app, then an API or MCP server, then the agent browser for everything else. The browser works on any website a person can use, so if the agent can’t reach something another way, it can still get the job done there with no custom build.

​

1. Connector

If Gumloop has a connector for the app, use it. Connectors call the app directly, so they’re the quickest way to read and update data. Gumloop has 300+ built-in connectors, including Gmail, Slack, Notion, Salesforce, and Google Docs. Example: “Summarize my unread Gmail from today.”

​

2. API or MCP server

If there’s no connector but the system has an MCP server, add it as a custom MCP server. See Custom, proxied, and hosted MCP. If it only has an API, store the key as a secret on the agent and ask the agent to save the calls as a skill. See Call integrations from a skill. Example: “Look up this order in our internal order service.”

​

3. Agent browser

For everything else, turn on the agent browser. See Agent browser. The agent opens the site and clicks, types, signs in, fills forms, and downloads files the way you would. It’s the right choice when:

• The app has no connector, API, or MCP server. Vendor portals, supplier and partner sites, government and bank websites, older internal tools.
• The connector or API doesn’t cover the action. Some settings, reports, and exports only exist in an app’s website.
• You’d otherwise have to build something. No API key, integration, or code to set up. Turn on the browser and describe the task.
• You need to see the page. Check how a site looks on desktop or mobile, or confirm what a customer sees.

Example: “Sign in to our vendor portal and download last month’s invoice.”

​

Why did my agent use the browser instead of my connector?

The agent only prefers connectors that are added to that agent. A connector you connected in Settings but didn’t add to the agent doesn’t count, so a browser-enabled agent may open the website instead and ask you to sign in. To fix it, add the connector to the agent: open the agent, go to the Agent tab, expand Connectors, and click Add. See Give an agent a connector. Then run the task again. If the connector is already on the agent, check that its account is signed in and that the tool the agent needs is turned on.

​

How do I make the agent use the option I want?

• Point at a connector: name it in your message, for example “Use the Google Docs connector to create the doc.” You can also pick Mention integration from the + menu in the message bar.
• Force the browser: ask for it, for example “Use your browser to open the vendor portal and download the invoice.”
• Keep it off the browser: say “Don’t use the browser for this.” To remove the option, turn the browser off in the agent’s Browser section and click Save.

One task can mix options, for example the browser to pull a report from a vendor portal and the Slack connector to post a summary. If you repeat it, ask the agent to “turn this into a skill.”

​

Related