Connect an AI (MCP)

Connect Claude, ChatGPT or any MCP client to FluenzR, with the list of tools and the approval setting.

Connect an AI (MCP)

The FluenzR connector uses the Model Context Protocol (MCP). Once connected, an AI assistant such as Claude or ChatGPT can read your FluenzR account and prepare your work: contacts, groups, campaigns, replies, meetings. The AI prepares, you decide what goes out.

The connector is included in the Pro plan, trial included. It is the same account and the same data as the app: nothing is copied elsewhere.

Server address

https://app.fluenzr.co/api/mcp

You sign in with your FluenzR account (OAuth 2.1): no API key to copy. The page Settings > AI > Connection in the app shows this address and the steps below.

Connect your AI

Claude.ai and Claude Desktop

Open Settings, then Connectors.
Click Add custom connector.
Paste the server address above.
Sign in to FluenzR when the authorization window opens, then click Authorize.

Claude Code

claude mcp add --transport http fluenzr https://app.fluenzr.co/api/mcp

In Claude Code, type /mcp, pick fluenzr and authenticate.

ChatGPT

In ChatGPT, open chatgpt.com/plugins, click Add (top right), then Add a custom MCP server.
Give it a name (for example FluenzR), paste the server address, keep authentication on OAuth, tick the confirmation box, then click Create as plugin.
Click Connect: the FluenzR page opens. Sign in (password or Google), then click Authorize.
In a new conversation, type @FluenzR and pick the plugin. Mention @FluenzR in every new conversation.

ChatGPT's interface changes often: if the labels differ, look for the option to add a custom MCP server. In ChatGPT's settings you can also ask it to confirm before each action.

Another MCP client

Any client that supports remote MCP servers over HTTP (Streamable HTTP) with OAuth can connect:

Add a remote MCP server and give it the address https://app.fluenzr.co/api/mcp.
The client registers itself (dynamic client registration) and opens the FluenzR sign-in page.
Sign in, read the permissions listed, then click Authorize. Deny cancels the connection.

What you authorize

The authorization window lists the permissions (scopes) requested by the AI:

Permission	What it allows
`mcp:read`	Read your contacts, campaigns, mailboxes and statistics
`mcp:contacts`	Create contacts and contact groups
`mcp:campaigns`	Create, start and pause email campaigns
`mcp:full`	Everything above (never ticked by default)

You can untick a permission before authorizing. Without mcp:campaigns, for example, the AI can read and import contacts but cannot create or start a campaign.

Tools

The AI sees 61 tools: 23 only read, 38 change something in your account. Every call to a write tool, successful or not, is logged in Settings > AI > Activity. Read tools are not logged.

Read tools (23)

They change nothing.

Tool	What it does
`get_account_status`	What is missing on the account before sending (subscription, mailbox, DNS, signature, address...), in order
`get_onboarding_questions`	The setup questions the AI asks you
`get_campaign_brief`	The brief for a new campaign: what the account already knows, open questions, rules
`get_mailbox_connection_link`	A link to connect a mailbox. The password is typed by you on the page, never in the chat
`list_email_accounts`	Your sending mailboxes with their limits
`get_mailbox_health`	Health report of one or all mailboxes
`list_contacts`	Search and filter contacts
`get_contact`	One contact in detail
`get_contact_history`	Everything that happened with one contact
`list_contact_groups`	Your contact groups
`list_group_members`	The contacts of a group
`list_campaigns`	Campaigns with status and basic stats
`get_campaign`	One campaign: sequence, mailboxes, groups, counters
`get_campaign_stats`	Statistics of a campaign, step by step
`preview_email`	An email step as it will be sent, variables filled. Nothing is sent
`get_dashboard_stats`	Dashboard overview
`get_inbox_counters`	Inbox counters by status and category
`list_inbox_conversations`	Inbox conversations, email and LinkedIn merged
`list_replies`	Contacts who replied
`get_conversation`	The whole conversation with one contact
`list_meetings`	Upcoming, past or canceled meetings
`get_booking_link`	Your booking link
`list_call_tasks`	The calls and manual tasks waiting for you

Write tools (38)

They create or change something in your account. Some can send real messages (marked "sends" below).

Contacts and groups (permission mcp:contacts)

Tool	What it does
`create_contact`	Create a contact
`bulk_create_contacts`	Create up to 200 contacts in one call
`create_contact_group`	Create a contact group
`update_contact`	Update fields and tags of a contact
`remove_contact_from_group`	Take a contact out of its group (the contact is kept)
`pause_contact`	Exclude a contact from all campaigns until resumed
`resume_contact`	Resume a paused contact
`stop_contact_in_campaign`	Stop sending to one contact in one campaign (cannot be resumed)
`mark_contact_bounced`	Record a bounce seen by hand
`unsubscribe_contact`	Unsubscribe a contact who asked to stop (cannot be undone)

Replies, meetings and calls (permission mcp:contacts)

Tool	What it does
`set_reply_status`	Classify a reply in the Inbox (to handle, interested, meeting, not now, not interested)
`draft_reply`	Write a draft answer with the reply assistant. Nothing is sent. Costs 1 AI credit
`send_reply`	Sends. Answer a contact in the same email thread
`create_meeting`	Create a meeting linked to a contact. Sends an invitation if asked
`complete_call_task`	Complete a call or manual task with its result
`log_call`	Log a phone call on a contact

Account settings (permissions mcp:campaigns and mcp:contacts)

Tool	What it does
`update_sender_settings`	Signature, postal address, personal-data page links, account language
`update_brand_profile`	Brand profile used to write sequences (sector, tone, persona, offer...)
`update_booking_page`	Set up the public booking page
`start_warming`	Start the warming of a custom SMTP mailbox

Campaigns (permission mcp:campaigns)

Tool	What it does
`create_campaign`	Create a campaign as a draft (email sequence or empty multichannel frame)
`add_email_step`	Add an email step
`add_linkedin_step`	Add a LinkedIn step (needs a connected LinkedIn account)
`add_wait_step`	Add a wait (up to 90 days)
`add_condition_step`	Add a yes/no condition
`update_campaign`	Change name, description, sending days, language, mailboxes, groups
`update_campaign_step`	Edit a step. On a running or paused campaign, only the text can change
`remove_campaign_step`	Remove a step from a draft
`add_contacts_to_campaign`	Add contacts or groups to a campaign
`send_test_email`	Sends. One test email, to your own address only
`start_campaign`	Sends. Launch a campaign
`resume_campaign`	Sends. Resume a paused campaign
`pause_campaign`	Pause a running campaign
`archive_campaign`	Archive a campaign (history kept)
`delete_campaign`	Delete a draft that never sent anything (cannot be undone)

LinkedIn sourcing (permissions mcp:contacts and mcp:campaigns)

Tool	What it does
`search_linkedin_people`	Search LinkedIn people through your connected account. Nothing is saved
`add_linkedin_results_to_group`	Add chosen profiles to a contact group
`attach_linkedin_search_to_campaign`	Feed a draft campaign with a continuous LinkedIn search

Safeguards stay active whatever the AI asks: a campaign is refused when it would send nothing (no audience, no usable mailbox), when a sending domain has no SPF or DMARC, or when an email has more than 2 links or a URL shortener. The first 20 recipients go out alone and the campaign pauses if more than 10 % bounce.

Example requests

Check my account and tell me what is missing before I can send.

The AI calls get_account_status and lists the steps, with a direct link for those that need you.

List my mailboxes and tell me how many emails I can send per day.

Show me the replies marked "to handle" and draft an answer to the first one. Do not send anything.

Import these 40 contacts into a new group "Bakeries Lyon", then prepare a three-email campaign for this group. Show me the sequence before you create it.

Which of my campaigns has the best open rate this month? Give me the numbers per step.

Pause the campaign "Autumn offer" and tell me how many contacts are still waiting.

Search LinkedIn for founders of marketing agencies in France and add the first results to a group.

Starter prompt

Once the AI is connected, paste this message into the conversation:

You are connected to my FluenzR account. I want to do as little as possible: you handle everything. Start with get_account_status: do the AI steps yourself, and for the ones that need me, give me the direct link. Then ask me the questions from get_onboarding_questions, one or two at a time, to set up my account and my first campaign as well as possible. Before each new campaign, call get_campaign_brief and ask me the questions that are still open, two or three at a time, then restate the brief for my approval before create_campaign. Follow the advice in the responses: if a mailbox is new or warming up, start with very few sends. Never ask me for a password in the conversation. Show me the sequence before calling start_campaign, and every reply to a contact before send_reply.

The "Validate my AI's actions before sending" setting

This box is on the page Settings > AI > Connection and on the page Settings > AI > To approve.

Box	Effect
Ticked (default)	Every sending action of your connected AI waits for your approval on the To approve page: campaign launch, restart of a paused campaign, text change on a running campaign, reply to a prospect. Nothing is sent before you approve.
Unticked	The same actions go out right away. This is a choice only you can make.

What happens when it is ticked:

Launch or restart: start_campaign and resume_campaign put the campaign in "awaiting approval" (nothing is launched or restarted until you approve). You see the rendered emails and a sample of contacts on To approve, then approve or refuse. If you refuse, the campaign goes back to draft with your reason, which the AI can read.
Text change on a running campaign: sending is suspended until you compare the old and the new text and approve. Refusing restores the old text.
Insert a step in a running campaign: same approval. The AI can first ask for an impact preview (dryRun), which saves nothing.
Reply to a contact: send_reply saves a draft on To approve. You read it, edit it and send it, or refuse it. The AI cannot approve for itself.

Good to know:

The box is ticked by default, for new accounts and for accounts that never touched it. Accounts that unticked it themselves keep their choice.
It only concerns your AI. What you do yourself in FluenzR (launch, restart, reply) needs no extra approval.
When you untick it, FluenzR reminds you that what your AI sends goes out without your agreement. Server safeguards (targeting, limits, DNS) stay active.
The AI cannot change this setting. Only you can, from the app.