Skip to main content

Triggerbee MCP Server

Learn how to install and work with Triggerbee's MCP server through your favorite AI tool.

Triggerbee MCP

Triggerbee MCP lets an AI assistant work inside your Triggerbee account on your behalf: read your data, answer questions about it, perform tasks, and make changes without you leaving the assistant you already use.

What an MCP is

MCP stands for Model Context Protocol. It is an open standard for connecting an AI assistant to an outside product. Think of it as a standard socket: the assistant is the plug, the product exposes a socket, and once they are connected the assistant can actually do things in that product instead of only talking about it.

In practice you type something like "how did the newsletter pop-up do last month?" or "soften the headline on the welcome campaign", and your assistant calls Triggerbee to do it, and performs it.

How to reach it

Server URL

https://mcp.triggerbee.com/mcp

Transport

Streamable HTTP. A hosted server, nothing for you to install or run locally

Sign-in

OAuth 2.1 in the browser, with PKCE. No API keys for you to copy or store.

Authorization server

https://api-gw.triggerbee.com

Permissions

mcp:tools:read, mcp:tools:write, offline_access

Client registration

Automatic (dynamic client registration) — any compliant MCP client can connect without Triggerbee pre-registering it.

What it exposes

37 tools and one live account list

Documentation

https://docs.triggerbee.com/mcp

How to install it

Whichever client you use, it is the same three steps: add the URL, sign in to Triggerbee in the browser window that opens, approve the permissions. Below are the specifics per client.

Claude (desktop or web)

  1. Open Settings → Connectors.

  2. Choose Add custom connector.

  3. Paste https://mcp.triggerbee.com/mcp and save.

  4. Click Connect, then complete the Triggerbee sign-in and consent screen.

Claude Code

claude mcp add --transport http triggerbee https://mcp.triggerbee.com/mcp

Then run /mcp, pick triggerbee, and choose Authenticate.

Cursor, VS Code, Gemini and other MCP clients

Add the server to your client's MCP configuration:

{   "mcpServers": {     "triggerbee": {       "type": "http",       "url": "https://mcp.triggerbee.com/mcp"     }   } }

Your client will prompt you to sign in the first time it connects. Clients that cannot do OAuth in a browser cannot connect to Triggerbee MCP.

Signing in and approving access

The consent screen comes from Triggerbee itself, and it shows you:

  • the name and logo of the application asking for access;

  • a warning if the application registered itself and has not been verified by Triggerbee;

  • which Triggerbee login you are signed in as;

  • which of your accounts the application will be able to work with;

  • the exact permissions it is asking for.

Nothing is granted until you press Authorize. Cancelling returns you to the application with no access given.

What the permissions mean

Permission

What you see

What it allows

mcp:tools:read

Read your accounts, campaigns and analytics

Look things up and report on them.

mcp:tools:write

Make changes to your accounts and campaigns

Create, edit, copy, publish and unpublish.

offline_access

Stay connected without asking for approval each session

Keeps the connection alive between sessions instead of making you re-authorize every time.

Permissions cap what your assistant can do. An application you granted only read access cannot change anything, no matter what you ask it to. The consent screen also supports a separate Publish on-site campaigns permission (mcp:campaigns:publish) for applications that ask for publishing rights on their own.

What the agent can do

Tools marked write can change your data and perform tasks on your behalf; everything else only reads.

Finding the right account

The server is stateless, meaning there is no "current" account. So your assistant first maps whatever you call an account ("the Swedish shop", often your brand name) to the right one, and asks which you meant if several match. It can also list every account you have access to.

  • triggerbee_accounts_resolve: turn a name you used into an account

  • triggerbee_accounts_list_accessible: list every account you can reach

  • triggerbee_accounts_get: account setup: tier, currency, URLs, tracking and attribution flags, paused state

Looking at your campaigns

  • triggerbee_onsite_campaigns_list: search campaigns by tag, type, status, name or expiry

  • triggerbee_onsite_campaigns_count: Can answer questions like "how many surveys do I have?"

  • triggerbee_onsite_campaigns_get: Get one campaign's name, type, status and scheduling

  • triggerbee_onsite_campaigns_get_settings: the full settings of a given campaign: scheduling, triggers, URL rules, repetition, tags, widgets and their targeting

  • triggerbee_onsite_campaigns_get_text: every piece of visible copy in the campaign as an editable list

Measuring your campaigns

  • triggerbee_onsite_campaigns_summary: impressions, clicks, conversions and revenue for one campaign

  • triggerbee_onsite_campaigns_trend: the same numbers over time: "is it improving?"

  • triggerbee_onsite_campaigns_widget_breakdown: per-widget performance inside one campaign

  • triggerbee_top_campaigns: your best performers, ranked by attributed revenue

Site-wide analytics

Five query tools share one shape: pick metrics, optionally group by something (device, country, source…), optionally add a time axis. so you can get almost any "X by Y over Z" answer without a purpose-built report.

  • triggerbee_traffic_query: sessions, pageviews, bounce rate, identified visitors

  • triggerbee_commerce_query: your site's own purchase revenue

  • triggerbee_campaign_query: widget engagement and campaign-attributed revenue across all your campaigns

  • triggerbee_signup_query: newsletter and contact signups

  • triggerbee_feedback_query: your NPS and CSAT results

  • triggerbee_site_overview: the dashboard KPI set in one call

  • triggerbee_goals_list: the conversion goals your numbers are based on

  • triggerbee_benchmarks_get: platform averages, to put your own numbers in context

Building and editing campaigns (write)

  • triggerbee_campaign_templates_list: browse the template catalogue by objective and tactic

  • triggerbee_onsite_campaigns_create_from_template (write): start a new draft from a template, automatically styled with your account's own brand settings

  • triggerbee_onsite_campaigns_update_text (write): rewrite the copy: "shorten the headline", "emphasise the 10% off"

  • triggerbee_onsite_campaigns_update_settings (write): name, scheduling, triggers, URL rules, repetition, tags, widget names

  • triggerbee_onsite_campaigns_update_targeting (write): who the campaign is shown to: audience and per-widget display conditions

  • triggerbee_onsite_campaigns_copy (write): duplicate a campaign, or copy it to up to 10 of your other accounts

  • triggerbee_widgets_duplicate (write): copy a widget, e.g. to start a desktop/mobile pair

Going live (write)

  • triggerbee_onsite_campaigns_publish (write): Publish a campaign

  • triggerbee_onsite_campaigns_unpublish (write): Unpublish a campaign

Connecting forms to email and CRM (write)

Connecting a form to a provider such as Voyado, Brevo, Dotdigital, Rule or Omnisend is always a preview-then-do pair: your assistant first shows you what would change, including how form fields would map and what the widget already pushes to, and only then makes the connection.

  • triggerbee_onsite_campaigns_plan_form_integration: preview the connection, changing nothing

  • triggerbee_onsite_campaigns_connect_form_integration (write): connect the form to the provider

  • triggerbee_onsite_campaigns_disconnect_form_integration (write): stop pushing to that provider

  • triggerbee_integrations_list: what is connected to your account

Audiences (write)

  • triggerbee_audiences_list: your existing audiences and their definitions

  • triggerbee_audiences_generate (write): build a new audience from plain language ("visitors who signed up and are in Sweden"). Triggerbee grounds your description in the account's real goals, tags, countries and segments, validates it and estimates its reach.

Help and diagnostics

  • triggerbee_help_search: searches Triggerbee's help centre, so "how do I…" answers cite real articles instead of guesswork

  • triggerbee_ping: connection check

Guardrails

These are enforced by the server and by the instructions it gives your assistant, not left to the assistant's judgement:

  • Edits are staged, never silently live. Your site keeps serving the last published version until a publish is explicitly requested. Your assistant is instructed not to chain a publish onto an edit without asking you.

  • Going live needs a clear yes from you. Publishing and unpublishing require explicit confirmation, and both are reversible.

  • No guessed accounts or IDs. Every account-specific answer names the account it came from, and IDs always come from a lookup and is never invented, never carried from one of your accounts to another.

  • A/B tests are off-limits. Publishing, copy edits and widget duplication are refused on A/B testing campaigns, which stay managed in the app.

  • Partial failures are reported. If a publish half-succeeds, you get the warnings verbatim instead of a clean-success story.

Everything is written to your activity log

Every change made through MCP is recorded in your account's activity log, tagged with the source MCP so you can tell it apart from changes made by a person in the app. Each entry keeps whose authorization was used, what was done (created, updated, deleted, published, unpublished, republished, invited), what it was done to, and which fields changed.

So your assistant's work is reviewable after the fact, exactly like a colleague's.

Good to know

  • One authorization can cover several of your Triggerbee accounts. The consent screen lists them before you approve.

  • Your assistant acts as you and your user. It can never do more than you could do yourself in the app.

  • Any MCP client can register itself, so only approve applications you recognize. Unverified ones are flagged on the consent screen.

  • The connection is a normal OAuth grant and you can revoke it; once revoked, that application's access stops working immediately.

Did this answer your question?