Fyreflock Try it free

Talk to Strangers from Claude Code: Fyreflock MCP Server

Updated 7 September 2026 · by Fyreflock

In short

Fyreflock has a remote MCP server at https://fyreflock.com/mcp. Connect it to Claude Code, Cursor or VS Code and your assistant becomes a courier: you type, it sends your words to a real person unchanged, and it shows you what comes back. You must be 18 or over, the same as meeting a stranger anywhere on Fyreflock. Unlike the website, messages sent this way pass through the Fyreflock server in plain text and through whichever AI provider you chose. Everyone on the other end is a real account that completed a camera check in a browser before it could be matched.

Key facts

What this actually is

Fyreflock pairs you with a stranger for a conversation. The website does that with video or text in a browser. This does the same thing through the assistant you already have open. It is text, with one exception: either of you can offer to move to a video call, which opens in a browser and never starts on its own.

Your assistant is a courier, not a participant. You type a line, it sends that line to the other person exactly as you wrote it, and it prints their reply back as a quote. It never answers on your behalf, never rephrases you, and never summarises them.

It works like texting rather than a phone call. Nothing can push a message into your terminal while you are idle, so replies arrive when your assistant next looks, or as a browser notification if you allowed those.

Which assistants this works with, compared

You need a Fyreflock account that has finished onboarding and completed the camera check. A terminal has no camera, so that step happens once in a browser and cannot be skipped.

In Settings, Connect an AI assistant, accept the terms and create a token. It is shown once and stored only as a hash, so copy it then.

Claude Code, Cursor and VS Code all accept a token in a header. Put it in an environment variable rather than in the command itself, so it stays out of your shell history:

export FYREFLOCK_TOKEN=your-token-here
claude mcp add --transport http fyreflock https://fyreflock.com/mcp -H "Authorization: Bearer $FYREFLOCK_TOKEN"

Then say find me someone and start typing. Say next for a different person and stop to leave. There are no slash commands: Claude Code takes those for itself before Fyreflock sees them.

claude.ai, Claude Desktop and ChatGPT are not ready yet. They want OAuth rather than a token in a header, and Fyreflock has not shipped that. This page will say otherwise when it changes.

Assistant host support, September 2026
HostStatusHow it connects
Claude CodeSupported and measuredToken in a header. Fyreflock holds a request open for ten minutes here, so waiting in the queue works properly
CursorWorks, not yet measuredToken in a header. Assumed to give up after 45 seconds, so the queue answers "still looking" sooner
VS CodeWorks, not yet measuredToken in a header, with the same caveat as Cursor
claude.ai and Claude DesktopNot yetWants OAuth, which Fyreflock has not shipped
ChatGPTNot yetWants OAuth; write tools also need a Business workspace

Turn off the approval prompt, or you will hate this

Most assistants ask permission before every tool call by default, which means confirming every single message you send. Answer "always allow" the first time Fyreflock asks. In Cursor, turn on per-tool auto-run for Fyreflock in your MCP settings.

A faster model also makes a real difference to how this feels. On the largest ones, every line you send carries a few extra seconds of thinking before it goes out.

What changes about your privacy

This is the part worth reading twice, because it is genuinely different from using the website.

Assistant messages pass through our server in plain text. Browser-to-browser chats travel directly between devices and we cannot read them. Relayed ones we can, and we hold a short buffer of recent lines so that a report can carry real evidence rather than an accusation.

They also pass through your AI provider. Everything you type and everything the stranger says goes through the model you chose, under that company’s terms rather than ours.

Use Fyreflock in a chat with no other connectors, files or tools attached. A stranger is typing into a context your assistant can read: whatever it can see, they can ask it about. Their words arrive wrapped and labelled as somebody else’s text for exactly that reason, and they are never instructions.

Research and agentic modes that browse or act on their own are a bad fit for the same reason. One conversation, one window, nothing else attached.

The rules, which are the same as everywhere else

18 and over, which is the same rule as meeting a stranger anywhere on Fyreflock. An account can be held from 16, and under 18 the website still has Roosts, but nothing here works at all: the token is refused before a single tool is offered.

Real accounts only. Everyone you meet completed a camera check in a browser before they could be matched. The honest description of the person on the other side is a verified account talking through an assistant, and they see the same description of you.

Word for word, and you can check. Your assistant is told to send exactly what you typed. The companion tab at fyreflock.com/app/live shows what our server actually sent and received in your name, rendered from our own record with the model nowhere in the path, so you never have to take its word for it.

Reporting, blocking, skipping and the house rules all work exactly as they do on the website, from tools rather than buttons. Assistant mode is free; a pass adds matching preferences and nothing else.

Frequently asked questions

Is this a chatbot?

No. There is a real person at the other end of every conversation, and their account completed a camera check in a browser before it could be matched. Your assistant only carries the words between you.

Can my AI assistant read my messages?

Yes, necessarily, because it is the thing sending and receiving them. So can whichever AI provider you use, under their terms. This is the main way assistant mode differs from the Fyreflock website, where messages travel directly between browsers and nobody in the middle can read them.

Does the other person know I am using an assistant?

Yes. Their peer card says so, and it is deliberately not hidden, because their words are passing through an AI provider they did not choose.

Will my assistant change what I write?

It is instructed not to, in the tool descriptions and again in every result it gets back. You can check independently in the companion tab, which shows what the Fyreflock server actually sent in your name, rendered from our own record rather than from the model.

How do I know when somebody replies?

Your assistant sees replies when it next looks, which is when you next type. For anything faster, allow notifications once from Settings and your browser will tell you a reply arrived even with every tab shut.

Does it cost anything?

No. Assistant mode is free, the same as the website. Paid passes only add matching preferences such as a gender filter, interest tags and queue priority.

Is there an age limit for assistant mode?

Yes, 18 and over, and it is the same limit as meeting a stranger on the Fyreflock website. You can hold an account from 16, and under 18 the website still has Roosts, but the assistant surface is closed to you completely.

Talk to someone real

Every face on Fyreflock passes a live camera check. No coins, 18+ only.

Start a conversation

Related