Connecting an assistant

Connecting is once per assistant, per person. After that the connection stays signed in and refreshes itself. Read AI assistants and the MCP server first if you have not, and confirm the four prerequisites in Turning MCP on before sending anyone through these steps.

The server address#

https://app.mymodero.com/api/mcp/mcp

That is the whole configuration. There is no API key, token, or secret to paste anywhere. The user’s client opens a browser window, they sign in to Modero, and they approve a consent screen. That is standard OAuth, and the AI vendor never receives their password.

Sign-in is served from app.mymodero.com even for a branded instance. Users on a branded instance should expect a Modero-branded sign-in window and use their normal credentials. Tell them in advance and it is a non-event.

Claude, web or desktop#

1. Settings, then Connectors.
2. Add custom connector.
3. Paste the server address. Leave Client ID and Client Secret blank: Modero registers the client automatically.
4. Connect, sign in, approve.

Claude Code#

claude mcp add --transport http modero https://app.mymodero.com/api/mcp/mcp

Then run /mcp and choose Authenticate. A browser window opens for sign-in.

ChatGPT and Gemini#

Add it as a custom connector or custom MCP server with the same address. Both vendors are on the auto-approval allowlist, so registration clears without a human.

Anything else#

Other clients register fine and are then held for review. The user is redirected to a /oauth/pending-approval page naming the client, and a Modero platform admin approves it. Locally-run tools using a 127.0.0.1 callback are approved without review. See Turning MCP on.

Confirming it actually worked#

A tool list on its own proves nothing. The list is published before any access check runs, so it looks identical for a user with full access and a user whose plan has no mcp_access at all.

Ask for something that needs live data instead. “How many services are in the inventory?” and then compare the number against the app. A real answer means the connection is good end to end.

What the user sees on the consent screen#

The consent screen names the scopes the client is asking for:

Scopes set a ceiling, never a floor. A connection granted mcp:write still cannot write anything the user’s own role does not allow. If a user only intends to ask questions, approving a read-only connection where their client offers the choice is good practice, and you can enforce it from your side as well. See Governing connected assistants.

Common questions#

Do users need an API key? No. Sign-in happens in the browser the first time, exactly like signing in to the app. There is nothing to copy, store, or rotate.

Why does a Vigilis user see a Modero sign-in page? Branded instances run on the Modero platform and the sign-in window is served from there. Expected. Normal credentials work.

What can they do once it is connected? See What an assistant can do for the tool catalog, and Getting good answers for how to ask. If the connection is not behaving, work down When MCP is not working.

One user’s connector shows up several times in the admin console. Is that a bug? No. Each registration mints a new client ID, and a single connector often registers many times behind the scenes. The console collapses them into one row per application, and actions fan out across the whole group.