Alpative MCP
Sign inGet started →FREN

Home · Guides

Connect Alpative to n8n, Make or a server-side agent with an agent key

Let an n8n or Make scenario, or an agent running on a server (OpenClaw, Hermes Agent, Claude Code, Cursor…), query your marketing data with no browser involved: create an agent key, put it in the right place, launch your first automatic report.

7 min readUpdated August 18, 2026
By the end of this guide
  • Know when an agent key is needed (and when it is not)
  • Create a key in the console and store it in the right place
  • Set up the MCP node in n8n and the MCP Client module in Make
  • Connect a server-side agent: OpenClaw, Hermes Agent, Claude Code, Cursor or your own framework
  • Run a weekly automatic report and fix the common errors

When to use an agent key

Claude, ChatGPT or Cursor connect on their own: you paste the Alpative address, click “Authorize”, done. On your own machine, they don't need a key.

An agent key is only for a program that has no browser to click with: an n8n or Make scenario, a script that runs every night, an agent running on a server (OpenClaw, Hermes Agent, Claude Code over SSH, Cursor on a remote machine, a custom bot). The key replaces the “Authorize” click. It gives that program exactly your rights: your connected accounts, your plan, your quota, nothing more.

1. Create the key in the console

01
Open the “Agent keys” cardIn your Alpative console (My account), scroll down to the “Agent keys” card.
02
Name the key after its useFor example “n8n Monday report” or “Make budget alerts”. You will know which one to revoke later.
03
Give it an expiry if you canThe expiry field is optional, but a key that expires limits the damage if it leaks. 90 days is a good setting.
04
Copy it right awayThe key starts with alp_live_ and is shown only once. If you lose it, revoke it and create another one.

The address to use in n8n or Make is your all-in-one MCP, the same one as for Claude or ChatGPT:

MCP address

https://mcp.alpative.com/marketing

2. In n8n

n8n has a dedicated node for MCP servers. The simplest setup is to plug it into an AI agent: you write the instruction in plain language, the agent picks the right Alpative tools and writes the answer.

01
Add an “AI Agent” nodeAttach a chat model to it (OpenAI, Anthropic, Google… with your own credentials). That model reads your data and writes the summary.
02
Add the “MCP Client Tool” under the agentIn the agent's “Tool” output, look for “MCP Client Tool”.
03
Fill in the address and the transportEndpoint: https://mcp.alpative.com/marketing. Server Transport: “HTTP Streamable”. On an older n8n version the field is called “SSE Endpoint”: update n8n, Alpative only serves the HTTP Streamable transport.
04
Pick “Bearer Auth” as authenticationCreate a Bearer Auth credential and paste your alp_live_ key in the Token field. n8n encrypts it; it is never written in clear text in the workflow.
05
Leave “Tools to Include” on “All”The agent sees all your Alpative tools. You can narrow it down later, for example Analytics only for a focused report.
06
TestOpen the AI Agent node's chat and ask a first question. If your list of properties comes back, everything is wired.
Test question

List my Google Analytics properties available through Alpative MCP.

Example: the Monday morning summary

A “Schedule Trigger” set to Monday 8 am, then the AI Agent with the instruction below, then a Gmail, Slack or Google Sheets node to deliver the result. Three nodes, a report that arrives on its own.

Agent instruction

On my Analytics property [my property name], compare the last 7 days with the previous week: sessions, users, conversions, main acquisition channels. Give the three key facts and one recommendation, 10 lines maximum, through Alpative MCP.

3. In Make

Make has an “MCP Client” app that loads the tool list of an MCP server and makes them usable like any other module.

01
Add a module from the “MCP Client” appTwo modules exist: “Call a tool” (you pick the tool and fill in its fields) and “Execute an action with AI” (you write an instruction, Make picks and chains the tools).
02
Create the connectionIn the MCP Server field, choose “+ New MCP Server”. URL: https://mcp.alpative.com/marketing. In “API Key / Access token”, paste your alp_live_ key. Make sends it as the access token on every call.
03
Pick the toolWith “Call a tool”, your Alpative tools appear (ga4_run_report, gsc_search_analytics, google_ads_run_report…) and the fields populate automatically: property, dates, metrics.
04
ChainAdd an Email, Slack or Google Sheets module to deliver or archive the result, and a scheduled trigger to run it every week.

4. OpenClaw, Hermes Agent and other agents

Any agent that can talk to a remote MCP server works the same way: the Alpative address, the HTTP Streamable transport, and the key in the Authorization header as a Bearer token. Three common examples, then the general rule.

OpenClaw

OpenClaw (self-hosted personal assistant, driven from WhatsApp, Telegram, Slack…) adds an MCP server in one command. Store the key in an environment variable first, then:

OpenClaw, command line

openclaw mcp add alpative \ --url https://mcp.alpative.com/marketing \ --transport streamable-http \ --header "Authorization: Bearer $ALPATIVE_KEY"

Hermes Agent

Hermes Agent (Nous Research) reads its MCP servers from ~/.hermes/config.yaml. Put the key in ~/.hermes/.env as ALPATIVE_KEY, then declare the server; HTTP Streamable is the default transport for a url entry:

Hermes Agent, ~/.hermes/config.yaml

mcp_servers: alpative: url: "https://mcp.alpative.com/marketing" headers: Authorization: "Bearer ${env:ALPATIVE_KEY}"

Then reload with /reload-mcp in Hermes: the Alpative tools show up in the agent's tool list.

Claude Code, Cursor and the rest

On your own machine, Claude Code and Cursor use the 2-click authorization (they open the browser). On a headless server, over SSH or in continuous integration, the key takes over. Claude Code, for example:

Claude Code (server, CI)

claude mcp add --transport http alpative https://mcp.alpative.com/marketing \ --header "Authorization: Bearer $ALPATIVE_KEY"

For a custom agent framework or another tool (Cursor, Cline, an MCP SDK), the rule is the same: HTTP Streamable transport, header Authorization: Bearer alp_live_…, header Accept: application/json, text/event-stream. No special initialization, no session to open.

Check and troubleshoot

  • 401 or “Unauthorized” error: the key is mispasted, revoked or expired. Check the “Agent keys” card in the console (the last-used date tells you whether it has ever worked) and create a new one if needed.
  • The tool answers that the source is not available on your plan: the scenario has your rights, no more. Google Ads, Meta or LinkedIn require the Pro plan, write actions the Expert plan.
  • Empty answer or “no property”: the source is not connected in the console. Connect it (“Connect” button); no change to the key is needed.
  • The quota is counted the same way as with Claude: one tool call = one request. A weekly report uses a few requests per week.
  • Quick test for a developer: a POST request to the MCP address with the header Authorization: Bearer alp_live_… and the header Accept: application/json, text/event-stream. The response comes back as text/event-stream (data: lines).

Stay in control

  • Store the key in n8n's encrypted credentials or in the Make connection, never in a text field of the scenario nor in a message.
  • One key per use: if a scenario is retired, revoke its key from the console. The effect is immediate.
  • Check the last-used date: a key being used while the scenario is stopped has leaked, revoke it.
  • The key manages neither your subscription nor your Google, Meta or LinkedIn connections: all of that stays in the console.

Ready to ask your data these questions?

Connect your accounts in two minutes, on the free plan.

Read next

Connect Alpative to Claude

4 min

The monthly client report

7 min

Reduce the token consumption of your MCP servers

9 min
Automate: n8n, Make, OpenClaw, Hermes · Alpative MCP