A Personal AI MCP Server for Your Conversations

Short answer: an MCP server is how an AI assistant reaches data that lives outside the chat window. Fluent runs one for your own conversations. Connect Claude, or any other client that speaks the Model Context Protocol, to https://mcp.becomefluent.io, sign in with your Fluent account, and the assistant can look up your recent conversations, pull up meeting notes and what you know about someone, and draft the follow-up. No API keys, no copy-paste. The steps for Claude, ChatGPT and Le Chat are on Connect your AI.

Here's the gap it closes. You open Claude to prep for a 2pm with a client. It can write a good agenda for a client meeting. It can't write one for this meeting, because it doesn't know that last Tuesday the client pushed back on the timeline, that you promised revised pricing by Friday, or that their VP joined the call and asked about security. That context exists. It's in a conversation you had out loud, and until now there was no way for your assistant to read it.

What MCP is, in one paragraph

The Model Context Protocol is an open standard, introduced by Anthropic, for connecting AI clients to tools and data. A server publishes a set of tools ("search conversations," "get meeting notes"). A client like Claude decides when to call them while it works on your request. Instead of pasting background into every chat, the assistant fetches the context it needs, when it needs it, from a source you've authorized.

Most MCP servers today connect developer tools: a code repository, a database, an issue tracker. A personal MCP server connects something stranger and more useful: your own working memory.

What Fluent's MCP server exposes

Fluent turns the conversations you capture into structured memory: meeting notes, people profiles that build up over time, and follow-ups. The MCP server makes that memory available to your assistant:

Memory and search: search_conversations finds conversations by what was said in them (transcripts and notes summaries), by title or by who was there, list_memories lists your conversations newest first, with filters for a date range and for a person who was there, and a cursor to page back, and get_memory returns the full transcript: what was said, by whom, and when.
People: list_people and get_person_context give the assistant the relationship context it has never had: what someone cares about, what you discussed last, what's still open between you. It can also set the notes on a person's profile.
Meeting notes: get_meeting_notes returns a conversation's summary, key points, decisions and action items, with the first 50 utterances of the transcript (about 30,000 characters at most); get_memory reads the rest of a long transcript page by page.
Six built-in workflows: prepare_for_meeting, draft_follow_up, weekly_review, visit_report, open_loops and decisions_since package the most common requests. "Prep me for my 2pm with David" starts from David's profile and the insights Fluent has built from your past conversations with him; the assistant can then search those conversations for anything else it needs.

Connect it in three steps

Capture conversations with Fluent. On your iPhone, through the phone's mic or a paired Limitless or Omi pendant. Nothing joins your calls; the audio is captured in the room. (More on that in AI meeting notes without a bot.)
Add the server to your AI client. Add https://mcp.becomefluent.io as a remote MCP connector. The setup guide walks through it.
Sign in and approve. The connection is OAuth 2.1: you sign in to Fluent and approve access. There's no key to paste into a config file, leak in a screenshot, or rotate later.

Then ask it something only your conversations could answer: "What did the client push back on last week?"

What it looks like in practice

Before a meeting. "Prep me for my 2pm with David." The assistant pulls David's profile and the insights from your past conversations, searches the recent ones, then writes talking points that start where you left off instead of from zero.

After a conversation. "Draft a follow-up from this morning's investor call." It reads the meeting notes, picks out the decisions and next steps, and drafts an email grounded in what was actually said. You edit and send it; it doesn't send anything on its own.

Across conversations. "What has Tomas brought up in our last few 1:1s?" The assistant finds the conversations Tomas was in, reads their transcripts, and pulls out what keeps coming up, with quotes and dates.

Why "personal" matters

Your assistant is already good at reasoning. What it lacks is your context, and that context is mostly spoken: decided in a 1:1, promised at the end of a call, mentioned once in a hallway. Notes apps capture the fraction you write down. Chat history captures what you typed into the chat. A personal MCP server is the first way to hand your assistant the part of your work that happens out loud.

It's also the opposite of lock-in. Your conversation memory isn't trapped in one app's interface: the same data answers questions in Claude today and in whatever MCP client you use next year.

Control and privacy

Scoped to your account. The client sees what your Fluent account can see, through the tools above, and nothing else.
You choose the clients. Only clients you add and sign in to can connect. To disconnect one, remove the Fluent connector in that client.
No training. Fluent never trains models on your conversations, and your data is encrypted at rest. Once a response reaches a third-party client, that client's own data policy applies, so read it as you would before pasting a document in.
Consent. If you're capturing conversations with other people, tell them. Fluent's recording consent guide covers how.

Where to start

If you already use Fluent, connecting takes about a minute: follow the MCP setup guide. If you don't, start by capturing a week of meetings. The MCP server gets more useful with every conversation it can draw on, and the first time your assistant reminds you of something you said and then forgot, you'll stop pasting context into chat windows.

Frequently asked questions

What is a personal AI MCP server?

MCP (Model Context Protocol) is an open standard that lets an AI client call outside tools and read outside data. A personal MCP server exposes your own data, not a company's, to the AI tools you already use. Fluent's MCP server exposes your conversation history, meeting notes, and people profiles.

How do I connect Claude to my conversations?

Capture conversations with Fluent on your iPhone, by phone mic or a paired Limitless or Omi pendant. Then add the Fluent MCP server (https://mcp.becomefluent.io) as a custom connector in Claude (any plan, including Free, which allows one custom connector) and sign in with your Fluent account. The connection uses OAuth 2.1, so there's no API key to copy or rotate.

Can ChatGPT use my conversations too?

Yes, on the web. ChatGPT can add a custom MCP server from its settings, though not on every plan, so check yours. The Fluent side is the same OAuth sign-in either way. Le Chat works too, including its free plan, and so does Claude Code for developers.

What can the AI see, and can I turn it off?

The client sees what your Fluent account can see, through the tools the server exposes, and nothing outside your account. To turn it off, remove the Fluent connector in your AI client. Fluent doesn't have a per-client switch to cut off a connected client yet.

Does the AI model train on my conversations?

Fluent never uses your conversations to train models. When you connect a third-party client, what that client does with the responses it receives is governed by that client's own data policy, so check it the same way you would before pasting a document into it.