Connect ChatGPT or Claude
mxAURA is an MCP server. Connect it once and your assistant can read your hubs and pools, write articles in your brand, place and publish them, and compose email sequences that HubSpot sends.
Where: Account, API and MCP tab (owners and admins). It has the API keys, the server URL with a copy button, step-by-step instructions for Claude, Claude Code, ChatGPT and Cursor, the highest access allowed, and the assistants connected to the account.
Claude: Settings, Connectors, Add custom connector, URL https://mcp.mxaura.ai/mcp, Connect. Sign in to mxAURA, pick the access level, Authorize. In a chat, switch the connector on from the tools menu.
ChatGPT: Settings, Connectors, Create (developer mode or a Business or Enterprise workspace). Same URL, authentication OAuth, leave client ID and secret blank, Connect, sign in, pick the access level, Authorize.
Claude Code: claude mcp add --transport http mxaura https://mcp.mxaura.ai/mcp, then /mcp to sign in. Or add --header "Authorization: Bearer YOUR_API_KEY" to use an API key from Account, API and MCP.
Access levels
You choose on the approval screen; the default is Read only.
- Read only: reads everything, changes nothing.
- Content: articles, hubs, AI drafts and generation, publishing, media.
- Campaigns: messages, pools, campaigns, blocks, flows, and reviewing or drafting messages with AI.
- Full: everything, including enrolling contacts (real email). Shown with a warning.
Deleting is never part of a level; it needs an API key with a delete scope. The assistant only sees the tools its level allows.
Highest access allowed
The account owner sets the most any assistant may be given (Account, API and MCP, Highest access allowed). It also holds an API key when the key is used through the MCP; the key keeps its own access on the API itself. The approval screen offers nothing above it, and lowering it cuts every connection down at once; connections above it are revoked. Content and Campaigns are separate: allowing one does not allow the other.
Asking for more when it is needed
An assistant can start with Read only. When it tries something that needs more (for example publishing), mxAURA tells it exactly which permission is missing; your assistant asks you to approve it, with the smallest level that covers it already selected. Assistants that cannot ask again show the message instead: reconnect and pick the level.
Connected assistants and keys
Connected assistants lists each assistant that signed in, who approved it, its level, and when it was connected and last used. Revoke ends its access within seconds; Revoke all ends every connection at once. To give one a different level, revoke it and connect again (or approve more when it asks).
An assistant set up with an API key (FormWise, for example) is not a connection: it uses the key. The API keys table shows it under the key's Last used, as "MCP (FormWise)" with the date. Revoking the key ends its access on the API and the MCP alike.
What it will and will not do
It confirms before publishing (public), unpublishing, deleting and enrolling a contact (real email). AI drafts it starts for articles that already have content wait as review drafts until applied. It never sees your prompt packs: they run inside mxAURA and are never returned. Every action shows in Activity, and AI work spends the account's AI budget.
Full detail and the tool list: developers.mxaura.ai.