Skip to main content
Connecting a server takes a Rippit Full Admin. Open Connections in Rippit and choose the MCP Servers tab. Some providers need work in their own admin console before Rippit can connect. Check your connector’s page first, since those steps are the ones that take time.

The flow

  1. Pick the provider, or choose Connect a custom server and paste an HTTPS MCP server URL.
  2. Choose how Rippit connects, if the provider offers a choice. See Identity below.
  3. Choose who can use it: only you, or everyone in your organization. See Scope below, which is the decision most worth getting right.
  4. Authorize. OAuth connectors send you to the provider to sign in and approve. Token-based connectors save immediately.
  5. Rippit reads the server’s tool list and enables the tools the server declares read-only.
  6. Review the tool selection. Once the connection verifies, Rippit opens the tool setup dialog before you continue, so you can turn on anything beyond the read-only default while you are still in the flow. You can reopen it later with Manage tools.
The connection is live as soon as it is saved. The next agent run picks it up.

Identity: who the agent acts as

MCP servers authenticate one of two ways, and the choice determines whose data the agent sees. Where a provider supports an application token, that is usually the better choice for a connection a team will rely on. A token does not carry one person’s private content into shared runs, and it does not break when that person changes roles or leaves. Intercom is the clearest example. Most providers only offer OAuth. In that case the connection acts as you, and the scope choice below decides who else gets to use that access.

Scope: who can use the connection

Every connection is created with one of two scopes. The last row is the one to read twice. Both scopes run on the identity that authorized the connection. Sharing a connection organization-wide does not give the agent your teammates’ permissions; it gives your teammates the agent’s access to your account, including private content, and any write tool you enable acts as you. Rippit asks you to acknowledge that explicitly before saving an organization-wide connection. Choose “only me” while you are trying something out. Choose organization-wide when the connection needs to serve a team, or when an Agent App has to run unattended, since those never see a personal connection.

When the provider needs setup work first

Rippit has to identify itself as an OAuth client to your provider. There are three ways that happens, and they differ in how much work lands on you. Nothing to do. Slack, Notion, Linear, Guru, Jira and Confluence, Hex, Replit, and Glean either recognize Rippit already or let it register itself automatically. Click connect and sign in. Google Drive also belongs here: it uses Rippit’s own Google app, so there is no Google Cloud project or OAuth client to create, and the only extra step is choosing items through Google Picker after sign-in. Your own OAuth client. Salesforce, Snowflake, Databricks, and HubSpot do not allow either, so an admin creates an OAuth client in that provider’s console and pastes its client ID (and secret, where required) into Rippit once. HubSpot calls it an MCP auth app, and it needs both the client ID and the secret. Rippit shows the exact authorized redirect URI to add to that client. Copy it from the dialog rather than typing it, since it must match exactly. A URL that is specific to your account. Snowflake, Databricks, Hex, and Glean do not have one shared endpoint. You supply the MCP server URL from your own account, and Rippit checks it points at a real host for that provider. Glean’s URL is in your Glean account settings, under Third party apps and MCP. Every connector’s dialog links to the provider’s own setup guide; open it before starting.

Choosing tools

Connections start read-only. Rippit reads the server’s tool list and enables only the tools the server itself declares read-only, so nothing that writes is on by default. From the connection’s card:
  • Manage tools lists everything the server exposes and lets you turn tools on or off. Tools are grouped by access level: Read, Write, Destructive, and Other, each showing how many are selected. Select all applies across every group.
  • Other holds tools the server did not annotate at all. Rippit cannot tell whether they write, so it treats them as writes and leaves them off until you decide.
  • Reset to read-only re-reads the server’s tool list and returns to the read-only default. It is also the fastest way to pick up tools a provider has added since you connected, because Rippit refreshes the list on connect, reconnect, reset, and when you open the tool picker, not on every run. A tool the provider adds later is never called until you refresh the list.
Two things worth knowing:
  • The agent can call any enabled tool without asking. Enable a write tool only when you want that to happen unattended.
  • Fewer tools tend to work better. A server with eighty tools spends the agent’s attention on choosing among them. Enable what your use case needs.

Keeping a connection healthy

A connection card shows Reconnect needed when its credential can no longer be refreshed, usually because access was revoked on the provider’s side, a password or token changed, or a long-lived grant expired. Click Reconnect and authorize again. The tool selection is preserved. Runs continue to work while one connection is unhealthy. That server’s tools simply are not offered.

Disconnecting

Disconnect removes the connection and its stored credentials, and the agent loses those tools for everyone the connection was shared with. Connecting again later is a fresh authorization. Disconnecting Slack is the one exception worth noting: it ends the MCP connection but leaves Rippit’s Slack app installed for report delivery. See Slack.