MCP server

AgentixSEO MCP server

Give Claude, Cursor or any MCP-capable assistant safe, scoped access to your AgentixSEO clients, audits, rankings and content tools.

Planned design

The MCP server described here is the planned design layered on top of the agency REST gateway, which is live today. Until the server ships, everything on this page can be done with the REST endpoints in the API section.

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external systems. An MCP server advertises a list of **tools**: named operations with typed inputs, and the assistant decides which to call while it works. The AgentixSEO MCP server is a thin translation layer: every tool maps one-to-one onto a read or an action that the agency REST gateway already exposes, so the permission model, per-client scoping and rate limits are identical.

What you can do with it

Ask questions about real data

"How did Northgate Dental rank last month?" resolves to live ranking rows, not a guess.

Run work end to end

Audit a site, generate an article from the winning keyword, then deploy it, in one conversation.

Build client-ready summaries

Pull rankings, backlinks and conversions, then have the assistant draft the monthly update.

Stay inside guardrails

A read-only key means the assistant can look at everything and change nothing.

How it fits together

Request path

AI client (Claude, Cursor, your own agent)
      │  MCP over stdio or HTTP
      ▼
AgentixSEO MCP server
      │  HTTPS + X-API-Key
      ▼
Agency REST gateway  ──►  scope check ──►  client scope check
      │
      ▼
Platform data and jobs (audits, rankings, content, outreach)

**Transport**: stdio for desktop clients, streamable HTTP for hosted agents.
**Auth**: the same agency API key you create under Agency → Automations → API. The server never sees a user password.
**Scoping**: the key's scopes decide whether write tools are advertised at all, and the key's client access decides which client_id values are accepted.
**Statelessness**: no data is cached in the server; every tool call is a fresh gateway request.

What it is not

It is not a second data store, there is one source of truth, the platform database.
It does not widen access. A tool that the key cannot authorise is not advertised to the assistant.
It does not auto-publish. Deploy and send tools are write-scoped and, by design, confirmed before they run.

Connect a client

Config snippets for Claude Desktop, Claude Code and Cursor.

Tool reference

Every tool, its inputs and the gateway call behind it.

Prompt recipes

Proven prompts for audits, content and reporting.

Safety and limits

Key scopes, confirmations, rate limits and logging.