What it does
Viva runs a remote Model Context Protocol (MCP) server. Once your assistant is connected, it talks to your Chief Marketing Agent: the AI marketing lead you named in Viva, which leads your agents and follows the rules you've taught it. You deal with that one persona, never with the agents one by one. For example:
- "Ask Jeeves how marketing is going and what I should focus on this week." (use whatever name you gave yours)
- "Tell Jeeves to pause prospecting until Monday."
- "What's waiting for my approval?"
- "Find wineries in Santa Clara County."
- "Which leads came in yesterday, and who hasn't been contacted?"
Questions and orders go to your Chief Marketing Agent through one tool, shown in your assistant as Ask plus the name you chose. Rename it in Viva any time; the new name shows up in your next chat. Quick lookups, like leads, the weekly brief and what's waiting for you, are read directly.
The assistant only acts on the Viva account you sign in with, and only within the permissions you approve. See every tool on the Tools page.
Connect
Add Viva to your assistant as a custom connector with this URL:
https://mcp.goviva.ai/mcp
- In your assistant, add a custom connector (Claude: Settings → Connectors → Add custom connector) and paste the URL. Leave any client ID and secret fields empty.
- Choose Connect. A Viva window opens. Sign in if you aren't already.
- Check the account name at the top, choose what the assistant may do, and select Allow.
Any member of a Viva account can connect, not only the owner. Step-by-step setup for each assistant is on the Clients page.
Permissions
The sign-in screen asks which of these the assistant may do. Reading is always included; the other two are off until you tick them.
| Permission | What it allows |
|---|---|
Readmcp:read | Look up leads, customers, appointments, agents, plans and the rest of your business data. Changes nothing. |
Make changesmcp:write | Add tags, update fields, decline a run, and let your Chief Marketing Agent act on your orders (turn agents on or off, run one). Without it, your Chief Marketing Agent can still answer questions but changes nothing. It cannot send email or texts. |
Approve sendingmcp:send | Approve an agent's run or follow-up draft. Approving can email real people from your account, so leave this off unless you want the assistant to do that. |
Sending an email or text message directly is never available to an assistant. Outbound messages always go through an agent's approval step.
Credits
Searching your business and the business directory is free. A few things can cost credits: a people search past the day's free allowance, running an agent, and approving a run whose workflow looks up email addresses. Talking to your Chief Marketing Agent costs the same as chatting with it in Viva: free on paid plans.
Those tools stop first and tell the assistant what it would cost. Nothing is spent until you agree, and the assistant passes confirm_spend: true. They are marked May use credits on the Tools page.
Disconnect
In Viva, open Settings → AI assistants. Every connected assistant is listed with the permissions it has. Select Disconnect and it loses access immediately.
To change an assistant's permissions, disconnect it and connect again.
API keys
For scripts, servers and clients that can't sign in through a browser, create an API key in Settings → AI assistants → API keys. Name it after what will use it, and list the tools it may call.
curl https://mcp.goviva.ai/mcp \
-H "Authorization: Bearer vsk_YOUR_KEY" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
A key can also be sent as X-Viva-Key. Treat it like a password: anyone holding it can act on your account within its tools. Delete it in the same place to revoke it.
Protocol & discovery
| Endpoint | https://mcp.goviva.ai/mcp (Streamable HTTP, POST) |
| Protocol versions | 2025-06-18, 2025-03-26, 2024-11-05 |
| Methods | initialize, tools/list, tools/call |
| Resource metadata | /.well-known/oauth-protected-resource/mcp |
| Authorization server | /.well-known/oauth-authorization-server (authorization code + PKCE, dynamic client registration) |
| Tool contract (JSON) | https://mcp.goviva.ai/mcp/contract |
Tool results carry both a text block and structuredContent. Each tool sets the standard readOnlyHint and destructiveHint annotations, so clients can ask before a change.