Claude Code & Codex

Both command-line agents speak MCP, so they can read your Dubsado account while you work.

Server URL

https://mcp.dubsado.dev/mcp

Claude Code

claude mcp add --transport http dubsado https://mcp.dubsado.dev/mcp

Then start Claude Code and run /mcp. Choose dubsado and pick the authenticate option. Your browser opens the Dubsado sign-in and consent screen. Once you approve, /mcp shows the server as connected and its tools become available.

Add --scope user to make the server available in every project instead of just the current one:

claude mcp add --transport http --scope user dubsado https://mcp.dubsado.dev/mcp

Useful follow-ups: claude mcp list to check status, and claude mcp remove dubsado to disconnect.

Codex CLI

Remote MCP servers are configured in ~/.codex/config.toml. Add:

[mcp_servers.dubsado]
url = "https://mcp.dubsado.dev/mcp"
auth = "oauth"

Then sign in. This opens the Dubsado consent screen in your browser:

codex mcp login dubsado

codex mcp list shows the configured servers.

If your Codex version predates remote-server support, bridge to it with mcp-remote, which runs the OAuth flow locally (requires Node.js):

[mcp_servers.dubsado]
command = "npx"
args = ["-y", "mcp-remote", "https://mcp.dubsado.dev/mcp"]

Any other MCP client

Configure the URL above as a remote (streamable HTTP) MCP server. Discovery is standard: an unauthenticated request returns 401 with a WWW-Authenticate header pointing at https://mcp.dubsado.dev/.well-known/oauth-protected-resource, which names the Dubsado authorization server. Clients that support OAuth 2.1 with PKCE and dynamic client registration need nothing else.

If something goes wrong

The browser window never opens: check that the URL you pasted ends in /mcp and starts with https://.
Approval succeeds but no tools appear: disconnect and reconnect the connector. Tool definitions are fetched fresh on each connection.
"Unauthorized" after it worked before: access was revoked or expired; reconnect from your AI app's connector settings.
An assistant reports partial results: it hit the 120 calls-per-minute limit. Ask again in a minute, or narrow the question.

← All set-up guides