Connections

A Connection is how an Agent reaches something outside Runva. Open Connections in the sidebar to see what you have and add more.

From the catalog

The quickest route is the built-in catalog. Choose a service, and Runva adds it and tests it in the same motion.

What's in there today: GitHub, Linear, Notion, Stripe, Sentry, Figma, Vercel, Hugging Face, Semgrep, DeepWiki, Context7, and Cloudflare Docs.

Adding one yourself

Add a connection takes anything that speaks MCP. You give it:

• Service and Name, which is how you'll recognise it.
• Acts as, which is the identity the other side sees.
• Server URL.
• Bearer token, if the server wants one.

The other kind is a service record, which holds an access token without bringing tools with it. That's what lets this Space's Computers push with git, for example.

Team or personal

Every Connection is one or the other, and it decides who can use it.

• Team means the Space owns it. Anyone's Agents can use it.
• Personal means it's yours. Only the Agents you own can use it, and nobody else's.

An Agent can use its owner's personal Connections. It can never use somebody else's.

What a connection is and isn't

A Connection grants access to one service. It doesn't authorize every action that service is capable of, and your approval settings still apply on top of it.

Credentials live in the Connection. They don't go into an Agent's instructions, into a Skill, into memory, or into a message. If you find yourself pasting a token into a conversation, that's the thing a Connection exists to stop.

Signing in by hand

Plenty of sites have no API worth connecting to. For those, take control of the Agent's Computer and sign in yourself. The browser session stays on that machine, so the Agent can carry on using it and won't have to ask again.