Connecting your AI tools

New to Connected AI? Start with What Connected AI is & how it works. This page is the practical half: getting each tool talking to your workspace.

Everything starts from Connections -> AI tools on your dashboard. The card there leads with Open the setup page, which is the quick way below and the one to take unless you have a reason not to. It also shows whether anything is connected yet, and when a connected tool last used your workspace.

Wiring a tool up by hand needs your connector address - e.g. https://insight.agentcroft.com/api/mcp. It's on the same card, under Add it by hand instead; the examples further down use that address.

The quick way: let your AI set itself up

There's a single setup page at insight.agentcroft.com/connect that both you and your AI can read.

Paste that link into whatever AI you're using and say "set this up". It works out which tool it's running in and takes it from there: Claude Code runs the command itself, Cursor writes its own config file, and a chat app like claude.ai or ChatGPT walks you through the few clicks it can't do for you.

It can't give itself access. Whatever it does, the connection only exists once you sign in and approve it - which is also where you choose what it's allowed to do.

The setup page is public and holds nothing secret: no token, no key, nothing about your workspace. It's safe to send to a colleague, and safe to paste into a chat. Your AI can also read it as plain text at /connect.txt.

Prefer to do it by hand? Every tool is covered below.

claude.ai & ChatGPT

These connect by signing in, so there's no token to create or store.

1. In the AI tool, go to its connector settings (in claude.ai: Settings -> Connectors -> Add custom connector).
2. Give it a name - "Marketing Brain" - and paste your connector address. Leave the advanced options alone: the tool detects how to sign in on its own.
3. Click Connect. You'll land on a page from the platform asking to approve the connection, listing exactly what it will be allowed to do.
4. Sign in with your workspace login if you're not already, review the permissions, and click Allow.

That's it - the connector shows as connected and your AI can use it straight away.

Sign in with the login you use for this workspace. If you're signed in with a different kind of account, the approval page says so and nothing is connected.

Each person connects their own AI with their own login, so activity is attributable and anyone can be disconnected individually from the card.

Claude Code

One command in your terminal - no token needed, because Claude Code can sign in the same way claude.ai does:

claude mcp add --transport http marketing-brain https://insight.agentcroft.com/api/mcp --scope user

Then run /mcp inside Claude Code, pick the connection and choose Authenticate. Sign in with your workspace login and click Allow.

marketing-brain is just the name the connection appears under - call it whatever you like. Leave off --scope user to add it to the current project only.

If you'd rather use a token, create one on the Connected AI card and add it as a header instead (replace YOUR_TOKEN):

claude mcp add --transport http marketing-brain https://insight.agentcroft.com/api/mcp --header "Authorization: Bearer YOUR_TOKEN"

Open a Claude Code session and run /mcp to confirm the connection is listed and healthy. Then just work: ask Claude to check your business context before drafting, and it will read who you are, your goals and your brand voice before writing a word.

Cursor

The fastest route is the Add to Cursor button on the setup page - one click and Cursor writes its own config. Otherwise, open Cursor's MCP settings and add a new server, or add this to your mcp.json (~/.cursor/mcp.json for every project, .cursor/mcp.json for one):

{
  "mcpServers": {
    "marketing-brain": {
      "url": "https://insight.agentcroft.com/api/mcp"
    }
  }
}

Then find the connection in Cursor's MCP settings and sign in. If your setup can't sign in, add a token instead:

{
  "mcpServers": {
    "marketing-brain": {
      "url": "https://insight.agentcroft.com/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Once saved, Cursor lists the connection and what it can do alongside its other tools.

Any other tool that accepts a remote MCP server with headers (VS Code and most coding agents) follows the same pattern: paste the address, and add an Authorization header with the value Bearer YOUR_TOKEN.

Agents you build yourselves

If your team builds agents on an AI provider's API, most providers now take remote MCP servers directly. On the Claude API, for example, the connection goes in the request's mcp_servers list:

"mcp_servers": [
  {
    "type": "url",
    "url": "https://insight.agentcroft.com/api/mcp",
    "name": "marketing-brain",
    "authorization_token": "YOUR_TOKEN"
  }
]

Check your provider's documentation for its current syntax - whatever the framework, the only two things it needs are the address and the token.

Keep tokens out of source code: load them from an environment variable or a secret store, and give each agent its own token so you can revoke one without breaking the others.

First things to try

• "Catch me up on what's changed since last Monday, then draft our stand-up note."
• "Get my business context, then draft three subject lines for our September newsletter."
• "How did marketing perform over the last 30 days? Include the caveats."
• "Read my latest monthly report and turn it into talking points for Monday's meeting."
• "Ask my marketing brain why TikTok underperformed last month." (Needs Read + ask.)
• "What open questions does my marketing brain have for me? Interview me, then save my answers." (Saving answers needs Collaborate - see Teaching the Brain through your AI.)

Ready-made recipes

Rather than remembering the right sequence each time, your workspace ships a handful of prompts - jobs you pick by name. In claude.ai they appear in the attachments or slash menu; in other tools, wherever that tool lists prompts.

You only see the recipes your connection can actually run, so nothing offers you a job that would be refused halfway through.

Attaching your brand guidelines

Some tools let you attach documents to a conversation. Yours offers two:

• Brand guidelines - your mission, positioning, who you sell to, brand voice and colours.
• Latest published report - the most recent report in full.

Attaching beats asking: pin the guidelines to a thread where you're writing ten emails and every draft starts in your voice, rather than hoping the AI checks first. If a document is empty (no brand identity filled in yet, or nothing published), you'll get a plain sentence saying so rather than a blank file.

If something's not working

Connected AI never fails silently: your AI tool always receives a message saying what's wrong and what to do next. Find yours below.

For anything not covered here, see Common issues & fixes - and the permissions, transparency and data-boundary background is in What Connected AI is & how it works and Security & privacy.