Connecting an AI client

Open a project, then open the Agent panel. Everything below starts there. Hosted clients connect from their own servers, so they need a Framulus that is reachable on the internet; local clients run on your machine and work against localhost fine.

Claude, as a custom connector

1. In the Agent panel, copy the link on the Claude row.
2. In Claude, open Connectors and add a custom connector.
3. Paste the link and approve the connection in the browser window that opens.

Nothing is installed on your computer. The row says so instead of offering a copy button when your Framulus is on localhost, because a hosted client cannot reach that.

Claude Code

claude mcp add --transport http framulus https://<your Framulus>/api/agent/mcp

Run it once, then approve the connection in the browser. Claude Code runs on your machine, so a local Framulus works.

Codex CLI

Open Advanced in the Agent panel, choose Connect Codex CLI, and paste the one-time setup into PowerShell. It installs the MCP entry, registers the local file companion if Node is present, and sends you to the browser to approve.

The local file companion

Uploading an image from your own disk needs a small local helper, because no image may travel inside an MCP call. Setup registers it as a second MCP server, framulus_files, which exposes one tool: upload_local_image. It reads the file, transfers the bytes to storage directly and hands back a ready asset id. Without it the agent can still import an image from a public https URL.

ChatGPT

The Framulus app for ChatGPT is not published yet, so there is nothing to install there. Codex CLI and Claude Code reach the same endpoint and the same tools in the meantime.