Quickstart

By the end of this you will have an MCP server running under Statio, a permission granted to yourself, and an agent calling a tool through the gateway with the credential injected server-side.

Prerequisites

A Statio account and an organization (app.statio.dev)
An MCP client: Claude Desktop, Cursor, Cline, Windsurf, VS Code, Zed
The Statio desktop app (statio.dev/download)

1. Install the desktop app

The desktop app is how your machine gets connected. It:

discovers which MCP clients you already have and which are configured,
writes the connection config for each one you pick: the right file, the right shape, no hand-edited JSON,
finds shadow MCP servers already configured on your machine, which is usually the first surprising thing you see.

You can configure a client by hand instead (see Connect a client), but the app is the supported path and it is the only one that tells you what was already there.

Note

Statio is a control plane, not an MCP client. The desktop app configures the clients you already use. It does not replace them.

2. Install a server from the catalog

In the dashboard, open Catalog and install one. Statio deploys it, discovers its tools, and registers each tool so it can be granted individually.

To bring your own instead, see Deploy your own MCP server.

3. Add the upstream credential

Open the server, go to Credentials, and add the key it needs upstream: a GitHub token, a Stripe key, whatever that API expects.

Warning

This is the last time you handle that value. It is encrypted at rest and injected into the upstream request at proxy time; it is never returned by the API and never sent to your agent.

4. Grant yourself the tool

Nothing is permitted by default. A freshly installed server is reachable by nobody, including you.

Under Permissions, grant a tool to yourself or a team. Grants are per tool, so a server exposing twelve tools can expose exactly the two you meant.

5. Connect your client

In the desktop app, pick the clients to configure and let it write the config. It sets the gateway URL and the credential for you.

If you would rather do it by hand, the shape is in Connect a client.

Note

Start a new session after connecting or installing a server. AI clients read their MCP servers once, when they start, so a session that was already open won't see the new tools. In Claude Code, run a fresh claude (resuming an old session keeps its old list); in Claude Desktop, Cursor and the others, quit and reopen.

6. Make a call

Ask your agent to use one of the granted tools. What happens:

The gateway authenticates the caller and resolves the principal.
It checks that principal holds permission for that tool.
It fetches the upstream credential from the vault and injects it.
It forwards the request and records what happened.

Open Audit in the dashboard. The call is there, with the tool, the principal and the outcome.

What to try next

Call a tool you did not grant. The gateway refuses it, and the refusal is recorded too. That is the whole product in one request: permission is checked per tool, not per server, and the answer is written down either way.