Connecting external AI agents (MCP server)

What it is

NexoFlow exposes an MCP (Model Context Protocol) endpoint, which lets an external AI assistant connect to your NexoFlow account and work with your content.

In practice: you can point an MCP-capable AI client at NexoFlow and then ask it things like "what's scheduled for next week?", "how is search traffic trending?" or "draft three blog posts about winter maintenance".

This is an advanced, optional feature. Most customers never need it- the built-in AI assistant (chat assistant) already does this inside the app.

Where to find it

Account Settings > MCP

Setting it up

1. Open Account Settings > MCP. 2. Copy the server endpoint. 3. Create a named personal token, selecting the minimum permissions the client needs. 4. Copy the token when it is shown- it is displayed once and cannot be retrieved later. Only a one-way digest of it is stored. 5. Configure your MCP client with the endpoint and the token as a bearer credential.

Tokens start with nxf_mcp_.

Exact client configuration varies by AI tool; the endpoint URL and the bearer token are the parts that stay the same.

Permissions (scopes)

A new token starts with PROJECT DISCOVERY ONLY. Everything else must be granted explicitly when you create the token.

• projects:read- See your projects, their status and connections
• content:read- Content calendar, blog plans, post lists, SEO details
• content:read_full- Full article bodies. Granted separately from content:read, because article bodies can quote your uploaded knowledge documents word for word.
• analytics:read- AI usage, social engagement, search keywords and search performance
• reviews:read- Google Business Profile reviews

You can also scope a token to specific projects, so an agent working on one client cannot reach another.

Safety: what an agent cannot do

The surface is deliberately read-first and conservative.

NEVER EXPOSED, under any configuration:

• Immediate publishing. There is no publish-now path at all.
• Deletion of anything.
• Your OAuth credentials for connected accounts.
• Your Developer API keys.
• Arbitrary access to the rest of the app.

WRITE ACCESS, when enabled: Write tools are off by default and must be turned on. Even then:

• Content reaches a live channel only through SCHEDULING, which leaves you a cancel window. A request to publish schedules 30 minutes ahead; unscheduling it in the dashboard is the cancel button.
• Drafting tools produce drafts a human reviews.
• An agent can DISABLE a social connection or PAUSE Autopilot, but never enable or resume them- authority only moves in the safe direction.

Every call is checked against your real permissions on the server. An agent cannot claim to be a different user or reach a project you do not have access to.

Actions taken through MCP are recorded.

A note on search data

Search keyword data includes the actual phrases strangers typed into Google. That text is third-party content, not instructions- a well-behaved agent should treat it as data. NexoFlow strips hidden and control characters from it before handing it over, but it is worth knowing where that text comes from.

Common questions