MCP Connections
Register custom HTTP MCP servers to extend Clevername with your own tools. Choose between always-on injection or plugin mode for fine-grained control over when tools are available.
What is an MCP connection?
An MCP connection is a custom HTTP MCP server that you register with Clevername. Unlike integration directory servers (which are pre-configured), connections let you point to any HTTP endpoint that speaks the MCP protocol — your own internal tools, self-hosted servers, or third-party services not yet in the integration directory.
Giving agents access
Agents never pick up a custom server on their own. If you're an admin (or on Free or Pro), connecting a server makes it available to your organization's agents; if you're not, tickAsk an admin to let our agents use this and the request goes to Org Review.
Then open the agent's Governance page and turn the tool on underConnected tools. Every action asks you first. The tool tells us what it can do; if it marks some actions as read-only and you trust those labels, you can let just those run without asking. Actions whose names suggest they change something always ask. Agents see the tools ascx_<server>__<tool>. Adding or editing a server, deleting one your agents use, and changing what an agent can use all need a session verified with two-factor.
If the tool changes an action the agent uses, that one action pauses until you review it. Actions the tool adds later stay off until you include them. Moving the server to a new address pauses all of it. If our security check holds an action back, Connected tools lists it as held for review; whoever connected the server can release it once the check clears it.
For servers that need sign-in you can use a static token, OAuth client credentials (your client secret is stored write-only), or keyless sign-in, where Clevername presents a short-lived signed token your server or Google Workload Identity Federation can verify, so no secret is stored at all.
Injection modes
| Mode | Behavior | Use when |
|---|---|---|
| Always-on | Tools injected into your own gateway sessions (IDE). Agents get a server's tools only through a grant (below). | Core tools you want available at all times (e.g., internal knowledge base) |
| Plugin | Tools available only when explicitly activated | Specialized tools you use occasionally (e.g., deployment scripts) |
Authentication
Connections support an optional Bearer token for authentication. The token is stored encrypted and sent as an Authorization: Bearer <token> header with every request to your server.
Open your custom servers
Navigate to Dashboard → Integrations. On the Tools tab, switch the view to Custom. You'll see a list of your existing custom connections, if any.
Add a new connection
Click + Add Custom Server. Fill in the required fields:
- Server Name — A display name for the connection (e.g., “Internal Wiki MCP”)
- Server URL — The HTTP endpoint of your MCP server (must be HTTPS in production)
- Auth Token (optional) — Bearer token if your server requires it
- Transport — Streamable HTTP (recommended) or SSE
- Trust Level — Sandboxed (rate-limited, no destructive tools), Trusted (full access), or Blocked
Choose the injection mode
Select Always-on to inject the server's tools into every gateway session, or Plugin to make tools available only when the server is explicitly activated for an agent or IDE session.
Test the connection
After saving, click the Ping button to verify that Clevername can reach your server. A successful ping confirms network connectivity and valid MCP handshake. If the ping fails, check that your server is reachable from the public internet and returns valid MCP responses.
Use the connection
Once connected, the server's tools are exposed to your agents and IDE through the MCP gateway — when activated (plugin mode) or automatically (always-on mode). Tools from custom connections are prefixed with the connection name to avoid conflicts with integration directory tools.