Skip to content

Connect MCP servers

Connecting an MCP server gives Ronja new tools — a Sentry, Notion, or internal-company server whose capabilities she can call mid-conversation, just like her built-in ones. (MCP is the Model Context Protocol, the open standard these servers speak.)

There is no “Add MCP server” form — like most resources, servers are registered only by asking Ronja:

  1. Tell Ronja in an exploration which server to connect — for example, “connect Sentry’s MCP server”.
  2. Authorize access when the card appears:
    • OAuth servers — the card reads “Ronja registered a client with this MCP server and pre-configured OAuth. Click Connect to authorize access.” Click Connect {name} and approve in the popup.
    • API-key servers — fill in the secure credentials card and click Save Credentials.
    • Public servers need no credentials at all.
  3. Watch the registration card flip to Connected — the server’s tools load on the agent’s next turn. If credentials are still missing, the card shows “Waiting for credentials” and the server activates automatically once the linked secret is filled in.

Every authenticated MCP server is linked to a secret; the server’s detail page shows it as Linked Secret. For OAuth servers, Ronja keeps the authorization fresh and installs it automatically on every call; for API-key servers, the key comes from the secret you filled in. Either way, credential values are never shown again after saving — and never exposed to the agent.

If the authorization expires, use the detail page’s Re-authorize action to run the consent popup again.

An MCP server is not shared with a checkbox — it inherits the scope of the feature that owns it. A server in a private feature is Personal (only you and the agent working for you use it); a server in a workspace- or organization-scoped feature is Shared with everyone who can access that feature.

Open the server from its feature hub. The detail page shows status chips (Live / Disabled / Error / Archived), Shared or Personal, the auth type (Public / API key / OAuth), and a tool count, plus the Endpoint URL and Allowed URL Prefixes — the hosts the server is allowed to be reached at.

  • Tools panel — lists every tool the server advertises, with a Refresh button to re-fetch the list. Tool lists are cached for about an hour and refreshed in the background. An amber Unusable section lists tools whose schemas aren’t compatible with the agent’s tool format — “They aren’t loaded into the agent. The server’s maintainer would need to fix the tool schema upstream.”
  • Enable / Disable — temporarily removes the server’s tools from the agent without deleting anything.
  • Delete — the confirm modal warns: “The agent will lose access to its tools.” Deletion is blocked while other resources structurally depend on the server.