Start

How to connect

The Mumega pot endpoint is https://mupot.mumega.com/mcp. Copy-paste setup for Claude Code, Cursor, and Codex, what ChatGPT can do today, and where your token comes from.

Not technical? Start with Mumega for business owners. This page is for anyone wiring an AI tool to a pot.

The endpoint

https://mupot.mumega.com/mcp

This is the one address every client connects to. It speaks MCP over streamable-HTTP. Opening it in a browser or calling it without credentials returns a 401. That is expected: it needs a bearer token or an OAuth sign-in. If a consent screen opens, it must be on mupot.mumega.com. Do not approve one on any other domain.

Which token, and where it comes from

You need a minted, agent-bound token (it starts with mupot_).

  • Who can mint one: an admin of your pot. They open the pot dashboard, find the Connect card, and mint a token for your seat. The raw value is shown once, so copy it straight into a file.
  • If you have no pot yet: workspace activation is by approval. Start a conversation at mumega.com/onboarding. There is no instant sign-up and no self-serve checkout today.
  • Keep it out of chat and git. Save it to a file with mode 600 and load it into your shell from that file:
mkdir -p ~/.mumega && chmod 700 ~/.mumega
# paste the token into this file once, then:
chmod 600 ~/.mumega/mupot.token
export SEAT_TOKEN="$(tr -d '\r\n' < ~/.mumega/mupot.token)"

Every block below keeps the raw token out of config files, except claude mcp add --header, which stores the token in plaintext in ~/.claude.json (and shows it in the process list while it runs). Use the environment-variable forms unless you accept that.

Claude Code

Recommended: a project .mcp.json that reads the environment variable, so the token stays out of the file:

{
  "mcpServers": {
    "mupot": {
      "type": "http",
      "url": "https://mupot.mumega.com/mcp",
      "headers": {
        "Authorization": "Bearer ${SEAT_TOKEN}"
      }
    }
  }
}

Start claude in a shell where SEAT_TOKEN is set. Not installed yet? See Install a coding harness.

If you accept that the token is stored in ~/.claude.json and visible in the process list while the command runs, you can instead add the server from your shell. This is the one path that does not keep the token out of a config file:

claude mcp add --transport http mupot https://mupot.mumega.com/mcp \
  --header "Authorization: Bearer $SEAT_TOKEN"

Cursor

Add this to .cursor/mcp.json in your project (or ~/.cursor/mcp.json for all projects):

{
  "mcpServers": {
    "mupot": {
      "url": "https://mupot.mumega.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:SEAT_TOKEN}"
      }
    }
  }
}

Start Cursor from a shell where SEAT_TOKEN is set. This shape follows Cursor’s documented MCP config. We have not walked it end to end for every Cursor version, so tell us if it does not connect.

Codex

Add this to ~/.codex/config.toml:

[mcp_servers.mupot]
url = "https://mupot.mumega.com/mcp"
bearer_token_env_var = "SEAT_TOKEN"

Start codex in a shell where SEAT_TOKEN is set.

ChatGPT

ChatGPT is early access. The Mumega connector for ChatGPT is added through ChatGPT Developer Mode (Settings, then Connectors) and is not a paste-in-a-config step. Request access on the ChatGPT page and the connector link is sent to you. We have not verified every ChatGPT plan end to end, so treat this path as untested until you have the link.

Check that it works

In your tool, call boot_context (no arguments) and then orient. On a minted token, do not call connect: your identity is already on the token. If boot_context returns your agent and squad, you are connected.

If you see 401, the token is missing or wrong. If a tool returns 403, your token does not carry that permission, so ask your pot admin.

Where next

Last updated: Oct 23, 2026