Connect Claude web (claude.ai)

You can connect claude.ai (Claude in the browser) straight to AgentValet as a custom connector. Claude signs in through AgentValet’s own OAuth flow, you approve once, and a new agent is provisioned for that connector. There is no key file, no MCPB install, and no CLI.

This differs from the MCPB and CLI paths: those put a private key on your machine. The Claude-web path uses OAuth instead, so there is nothing to paste and nothing to store locally.

Before you start

Find your Owner ID: dashboard, Settings, the Owner ID field in the Developer panel. It is a UUID. Your connector URL is:

https://api.agentvalet.ai/mcp/v1/<owner-id>

Step 1: Add the custom connector in claude.ai

In claude.ai, open your connector settings (Settings, then Connectors) and choose to add a custom connector. (Anthropic occasionally relabels this screen; look for the option to add a connector by URL.)
Paste your connector URL: https://api.agentvalet.ai/mcp/v1/<owner-id>.
Leave the OAuth Client ID and Client Secret fields blank. Claude identifies itself to AgentValet by URL (a client-id metadata document), so there is no client ID to type and nothing to register by hand.
Save, then start the connection.

Claude opens AgentValet’s authorization page in a popup.

Sign in with your AgentValet account (the same login you use for the dashboard).
On the consent page, review what is being connected and click Approve.

Approving hands Claude a short-lived token bound to your AgentValet MCP endpoint. Claude never receives a signing key.

Step 3: Grant the new agent some access

Approving provisions a brand-new agent for this connector, and like every agent it starts deny-by-default with zero scopes. Until you grant it something, every call Claude makes comes back denied.

In the dashboard, open Agents. The new connector agent appears (its type is the OAuth-connector kind, not a key-based agent).
Open it and configure its permissions: tick the platforms and scopes you want Claude web to reach.

Give it only what this browser session needs. It has its own identity and its own audit trail, separate from your Desktop or CLI agents, so you can revoke it on its own later.

Step 4: Confirm it works

Back in claude.ai, ask Claude to list the platforms it can use through AgentValet. It should return exactly the platforms you just granted. An empty list means the agent has no grants yet (go back to Step 3); an error means the connector did not finish authorizing (remove it and redo Step 1).

Where this connector shows up

Custom connectors are stored at your Claude account level, so once you add AgentValet on claude.ai it is available anywhere that reads that connector list. Anthropic’s documentation lists custom remote-MCP connectors as supported on:

Claude on the web (claude.ai)
Claude Cowork
Claude Desktop (it appears under Settings, Connectors alongside any AgentValet MCPB extension; use either one, you do not need both)

You add it once, in one place, and it follows you across those surfaces.

Claude Code (the CLI) inherits your account-level connectors when you are signed in, so AgentValet shows up there too; you can also add it directly with claude mcp add. Note the trade-off: the MCPB and CLI paths give a Claude Desktop or Claude Code agent its own signing key, a stronger and separately-revocable identity than the shared account connector. Use the connector for a quick, no-install start; use MCPB or the CLI when you want a per-agent key and its own audit trail.

Claude Code on the web works as well. Anthropic’s public connector list does not name this surface, but we have tested it directly: adding AgentValet as a custom connector there runs the same OAuth approval and provisions a new agent, exactly as on claude.ai. If you do not see a custom-connector option in your build, it may not have reached your account yet.

On Claude’s Free plan you can have only one custom connector at a time; paid plans allow more. See Anthropic’s custom connectors guide for the current surface list and limits.

If something went wrong

The consent page hangs on “Checking your sign-in…” Refresh the popup. If it persists, the AgentValet proxy serving your endpoint is out of date; the fix shipped in proxy 0.12.1, and on the hosted service (api.agentvalet.ai) it is already live.
“invalid_client”, or the connector will not authorize. Re-check the connector URL. It must be the full https://api.agentvalet.ai/mcp/v1/<owner-id> with your real owner ID, and the OAuth client fields must be left blank.
Claude connects but every action is denied. Expected until you grant scopes; see Step 3. Destructive scopes may additionally need per-request approval.
You do not see the new agent in the dashboard. Make sure you completed the consent Approve. Closing the popup early cancels provisioning; redo Step 2.