Connect an AI assistant to Argora

Argora speaks MCP (Model Context Protocol)— the open standard AI assistants use to connect to apps. One server covers the whole suite, so a connected assistant can work across mail, drive, docs, sheets, slides, notes, planner, chat, meetings, forms, flow, sign, books, and send in a single conversation: “what's unpaid and more than 30 days old?”, “summarize this week's #ops channel and file the notes in Drive”, “draft the renewal invoice and send it for signature.”

The server address is https://mcp.argora.net/api/mcp — that URL is the whole setup. Access is per person: you sign in with your own Argora account and approve the connection before anything is shared, and the assistant can never do more in any app than your own role allows there.

Two things vary independently, and both are decided when you connect: which account the assistant acts as — whoever you sign in as — and which workspace it works in, which comes from the server URL. The plain address, https://mcp.argora.net/api/mcp, needs neither spelled out: it connects the account you sign in as, in your default workspace. That is all most people ever need.

If you're signed in to more than one Argora account, connecting stops and asks which one to use — a connection lasts until you revoke it, so it shouldn't quietly attach to whoever happened to be signed in last. You can also switch from the approval screen.

To work in a specific workspace instead of your default, name it on the URL: https://mcp.argora.net/api/mcp?org=your-workspace. (Clients that let you set request headers can send X-Argora-Org instead.)

You can have several connections at once — your own account and a client's, or one account in two workspaces. Each is a separate server on the assistant's side, with its own name, its own URL, and its own sign-in. In Claude Code:

claude mcp add --transport http argora https://mcp.argora.net/api/mcp
claude mcp add --transport http argora-acme https://mcp.argora.net/api/mcp?org=acme

Then authenticate them one at a time (/mcp, pick the server, sign in). When the second one sends you to Argora, choose the other account at the picker — that connection binds to it, and the first one stays exactly as it was. In Claude's connector UI the shape is the same: add a second custom connector, then sign in as the other account.

Give each one a distinct name and URL. Assistants tell servers apart by those, and the ?org= does double duty as the distinguishing part. Name them for the identity rather than the product — argora-work, argora-acme — because the tools are named identically across connections, so in a conversation the server name is the only thing that says which account is about to act.

Each connection is listed and revoked on its own under Connected apps at id.argora.net — for the account it belongs to. Disconnecting one leaves the others running.

Read— your mail and calendar, files and documents, chat history, tasks and plans, forms and responses, meeting rooms, signature envelopes, and the books: P&L, balance sheet, trial balance, AR/AP aging, invoices, bills, and transactions. The same data you see when you sign in, and no more.

Change — send and file mail, create and edit documents, spreadsheets, slides, and notes, upload and organize files, post to channels, create and move tasks, build forms and flows, schedule meetings, send envelopes for signature, and record accounting work: invoices, bills, payments, and journal entries.

Confirms before it destroys — deletes, share-link changes, sender revocations, and other one-way actions are gated: the assistant must re-issue the call with an explicit confirmation, so nothing irreversible happens as a side effect of a vague request.

Never — the assistant is not a way around your permissions. Every call is re-checked by the app that owns the data, using your identity, so admin-only actions stay admin-only and other workspaces stay invisible. Platform-operator powers, billing changes, and identity actions that mint or reveal credentials are excluded from the tool set entirely — those stay in the apps, done by a person.

Each tool call — successful or failed — is written to an append-only log: who it ran as, which key or client made it, what was asked, and what came back. A call that can't be logged doesn't run. Secret-shaped values are stripped before the log is written, so the record shows what was done without becoming a second copy of your data.

You can read it yourself at mcp.argora.net/audit — or just ask the assistant what it changed, and it will read the log back to you.

Something looks wrong, or a client won't connect? Email support@argora.net. Suspected key compromise goes to security@argora.net.