Use Fellix from Claude Code or Cursor

Claude Code and Cursor can use Fellix the same way ChatGPT and Claude do. Look up a client, check today's appointments, draft a reply or pull a report without leaving your terminal or editor. You add Fellix once, sign in once in the browser, and the assistant then works as you, within the permissions you approve.

Any member of an organization can connect. The address is the same for every assistant: https://api.fellix.ai/mcp.

Claude Code

1. In a terminal, run:

   claude mcp add --transport http fellix https://api.fellix.ai/mcp
   

   Add --scope user to have Fellix in every project, not only the current one.
2. Start Claude Code, type /mcp, choose fellix and choose to sign in. Your browser opens Fellix.
3. Sign in to Fellix if asked, choose the Organization and the Permissions, then choose Authorize. Go back to the terminal.

Fellix's consent page. It names the assistant that asks, here ChatGPT.

Fellix's three ready-made prompts appear in Claude Code as commands:

• /mcp__fellix__inbox_briefing: what moved in the inbox lately and what needs a person, without changing anything.
• /mcp__fellix__follow_up_warm_leads: open hot and warm leads that have gone quiet, each with a drafted follow-up. Nothing is sent until you approve it.
• /mcp__fellix__today_appointments: the day's appointments in the organization's time zone, with what still needs confirming.

Cursor

1. Add Fellix to ~/.cursor/mcp.json to have it in every project, or to .cursor/mcp.json inside one project:

   {
     "mcpServers": {
       "fellix": { "url": "https://api.fellix.ai/mcp" }
     }
   }
   

2. Cursor offers to sign in the first time it connects to Fellix. Accept, and your browser opens Fellix.
3. Choose the Organization and the Permissions, then choose Authorize.

Other assistants

Any assistant that supports remote MCP servers with sign-in (Streamable HTTP and OAuth) can connect with the address alone: https://api.fellix.ai/mcp. It sends you to the same Fellix consent page.

Try these

• "Find the client Maria Lopez in Fellix and summarise her last conversation."
• "List today's appointments in Fellix, in the shop's time zone."
• "Draft a reply to Maria Lopez's latest conversation. Do not send it."
• "Using Fellix, show last week's pipeline: won, lost and the win rate."

Asking before acting

Both tools ask before they use a tool by default, unless you allowed it:

• Claude Code asks before each Fellix tool until you allow it for the session or permanently.
• Cursor asks before each tool unless you have turned on running tools automatically.

Tools that send a message or change data reach real clients and real records. Keep approvals on for them, at least until you know how the assistant behaves.

With an API key instead of signing in

For a script or a shared machine with no person signed in, an owner can create a Static key under Settings → API keys → Create key, granting only the access that job needs. The assistant then uses the key instead of a sign-in.

• Claude Code:

  claude mcp add --transport http fellix https://api.fellix.ai/mcp \
    --header "Authorization: Bearer $FELLIX_API_KEY"
  

  Claude Code saves the header in its settings on this computer. Do not add --scope project with a key: that writes it into the project's .mcp.json, which is shared with everyone who has the project.
• Cursor: read the key from an environment variable rather than writing it into the file:

  {
    "mcpServers": {
      "fellix": {
        "url": "https://api.fellix.ai/mcp",
        "headers": { "Authorization": "Bearer ${env:FELLIX_API_KEY}" }
      }
    }
  }
  

What changes with a key:

• The assistant acts as the organization, not as a person. It reaches what the key's access covers across the whole organization, and what it sends is not recorded under a person's name.
• Tools that need a person are not offered: taking, letting go of or handing over a conversation, and most reports. Of the reports, a key gets only the pipeline, won and lost over time, and tags reports.
• The key lasts for the expiry chosen when it was created (30 days, 90 days, 1 year, a date, or never) or until an owner revokes it.

Good to know

• Under your name. With a sign-in, everything the assistant sends or changes is recorded as done by you.
• How long it lasts. A sign-in connection lasts up to 90 days and ends sooner if unused for 30 days. Then sign in again: /mcp in Claude Code, or the sign-in Cursor offers.
• One organization per connection. To switch, disconnect and connect again, choosing the other organization.
• Limit. You can have up to 10 connected apps in each organization.
• Disconnecting. Account menu → Account Settings → Connected apps → Disconnect. Removing the server from Claude Code (claude mcp remove fellix) or from mcp.json stops the assistant using it, but disconnect in Fellix too to end the access.

Troubleshooting

Related

• What an AI assistant can do in Fellix
• Manage connected AI assistants
• Connect Fellix to Claude
• Developers: MCP server reference and API keys