Guide · Modu over MCP

Connect AI assistants to Modu

Modu runs two MCP servers. Add them to Claude, ChatGPT, Cursor, VS Code, Codex, Windsurf or Grok, and set up Discord servers or answer tickets from the assistant you already use.

Modu’s MCP servers

Modu Server Setup
Build and change your Discord servers from your AI assistant. Each request runs the same Modu agent as the chat: it plans the change and shows the plan before anything risky happens.
https://modu.ubits.tech/mcp
- List your connected servers list_servers
- Start a new server: the steps to make an empty one and add Modu start_new_server
- Server overview: channels, roles and features get_server_overview
- Run a setup request in plain words run_setup
- Check a run: status, planned steps and their risk get_run
- Approve a plan with no destructive steps approve_run
- Cancel a run cancel_run
- Undo a run undo_run
Modu Tickets
Work your Discord ticket queue from your AI assistant. Replies are posted in the ticket channel as Modu, labelled with the app that sent them.
https://modu.ubits.tech/mcp/tickets
- List open or closed tickets list_tickets
- Read a ticket's latest messages get_ticket
- Reply in the ticket reply_ticket
- Claim a ticket claim_ticket
- Close a ticket close_ticket
Modu for Slack
Work in the Slack workspaces linked to your Modu plan from your AI assistant. Each request runs the same Modu agent as the Slack console, and plans that change Slack wait for approval.
https://modu.ubits.tech/mcp/slack
- List the workspaces linked to your Modu plan list_workspaces
- Workspace status and a few counts get_workspace_overview
- Ask Modu to do something, in plain words run_request
- Check a run: Modu's reply and the plan get_run
- Approve a plan with no dangerous steps approve_run
- Drop a plan that is waiting cancel_run
Modu for WhatsApp
Work with the WhatsApp numbers linked to your Modu plan from your AI assistant. Each request runs the same Modu agent as the WhatsApp console, and anything that sends or changes something waits for approval.
https://modu.ubits.tech/mcp/whatsapp
- List the numbers linked to your Modu plan list_numbers
- Number status and a few counts get_number_overview
- Ask Modu to do something, in plain words run_request
- Check a run: Modu's reply and the plan get_run
- Approve a plan with no dangerous steps approve_run
- Drop a plan that is waiting cancel_run

Sign-in

Two ways to let an AI app act as you. Both can be revoked at any time.

Most apps

Connect with Modu

Add a Modu server URL in your AI app.
The app opens a Modu sign-in page. Sign in with Discord, as you do on this site.
Modu’s consent page shows the app’s name, what it asks for and which Modu server it can reach. Choose Allow.

The app gets a Modu token (OAuth), never your Discord password or token. It lasts an hour and is renewed while you keep the app connected.

Apps without sign-in

Personal access token

For the xAI API and any client that can only send a fixed header, create a token on AI connections (/app/settings/ai). Pick its name, scopes and an optional expiry.

It starts with modu_pat_ and is shown once, so copy it then. Send it as Authorization: Bearer modu_pat_…. Keep it secret, like a password, and revoke it on the same page.

Per app

Set it up in your app

Pick your AI app. Each tab follows that app's own documentation, linked at the bottom of the steps. Add the Tickets server the same way with its own URL.

Claude (web and desktop)

Add Modu as a custom connector. A connector you add on claude.ai is set up through your Claude account, so it also shows up in Claude Desktop.

Pro or Max: open Customize > Connectors, then choose + Add > Add custom connector.
Name it Modu, paste the Server Setup URL and choose Continue. Keep the authentication settings Claude detects; Modu’s sign-in supports “Register automatically”. Choose Add.
Choose Connect, sign in to Modu and allow access. Then turn the connector on for a chat from the + button, under Connectors.
Add a second connector, Modu Tickets, with the Tickets URL, if you want the ticket tools too.
Team or Enterprise: an Owner adds the connector first in Organization settings > Connectors (Add > Custom > Web). Each member then connects to it from Customize > Connectors.

Good to know

Custom connectors work on the Free, Pro, Max, Team and Enterprise plans. Free is limited to one custom connector, so pick Server Setup or Tickets.
On Team and Enterprise only Owners and Primary Owners (on Enterprise, also a custom role with Manage access to Libraries) can add a custom connector.
On Team and Enterprise, leave Request headers empty: a token there is shared, so everyone using the connector would act as that one Modu account.

Checked 2026-10-23 against Claude Help Center: custom connectors.

Server Setup URL

https://modu.ubits.tech/mcp

Tickets URL

https://modu.ubits.tech/mcp/tickets

Claude Code

Add both servers from your terminal, then sign in from inside Claude Code. If you log in to Claude Code with a claude.ai account, connectors you added on claude.ai are available there too.

Run the two claude mcp add commands. --scope user makes Modu available in all your projects; leave it out for the current project only (local, the default), or use --scope project to share it through .mcp.json.
Start Claude Code and type /mcp. Pick modu, then follow the steps in your browser to sign in to Modu and allow access. From a shell, claude mcp login modu does the same.
No browser sign-in? Add the server with your personal access token in an Authorization header instead.

Good to know

The token variant stores the token's value in your Claude Code config. Revoke it on /app/settings/ai if that file is ever shared.

Checked 2026-10-23 against Claude Code docs: MCP.

Terminal

claude mcp add --transport http --scope user modu https://modu.ubits.tech/mcp
claude mcp add --transport http --scope user modu-tickets https://modu.ubits.tech/mcp/tickets

Inside Claude Code

/mcp

With a personal access token

claude mcp add --transport http --scope user modu https://modu.ubits.tech/mcp \
  --header "Authorization: Bearer $MODU_TOKEN"

OpenAI Codex

Codex reads MCP servers from ~/.codex/config.toml, which the Codex CLI, the IDE extension and the ChatGPT desktop app share. Add Modu with codex mcp add or edit the file.

Run codex mcp add for each server. When Codex detects that a server supports sign-in, it starts the browser sign-in for you.
To sign in again later, run codex mcp login modu.
Or edit ~/.codex/config.toml (or .codex/config.toml in a trusted project) and add the two tables below.
Using a personal access token instead? Put it in an environment variable such as MODU_TOKEN and add bearer_token_env_var to the table. Codex sends it as the Authorization bearer token.

Checked 2026-10-23 against Codex docs: MCP, codex mcp source.

Terminal

codex mcp add modu --url https://modu.ubits.tech/mcp
codex mcp add modu-tickets --url https://modu.ubits.tech/mcp/tickets
codex mcp login modu

~/.codex/config.toml

[mcp_servers.modu]
url = "https://modu.ubits.tech/mcp"

[mcp_servers.modu-tickets]
url = "https://modu.ubits.tech/mcp/tickets"

With a personal access token

export MODU_TOKEN="modu_pat_..."
codex mcp add modu --url https://modu.ubits.tech/mcp --bearer-token-env-var MODU_TOKEN

# or, in config.toml:
[mcp_servers.modu]
url = "https://modu.ubits.tech/mcp"
bearer_token_env_var = "MODU_TOKEN"

Cursor

Install Modu with one click, or add it to mcp.json. Cursor signs in to Modu with OAuth (authorization code with PKCE).

Choose Add to Cursor below. Cursor opens and asks to install the server.
Or create ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project) with the snippet.
When Cursor asks you to sign in, a Modu page opens. Sign in and allow access.
For a personal access token instead, add "headers": { "Authorization": "Bearer ${env:MODU_TOKEN}" } to the server.

Add Server Setup to Cursor Add Tickets to Cursor

Good to know

The Add to Cursor links open the Cursor app, so they work on a computer with Cursor installed.

Checked 2026-10-23 against Cursor docs: MCP, Cursor docs: install links.

~/.cursor/mcp.json

{
  "mcpServers": {
    "modu": {
      "url": "https://modu.ubits.tech/mcp"
    },
    "modu-tickets": {
      "url": "https://modu.ubits.tech/mcp/tickets"
    }
  }
}

VS Code with GitHub Copilot

Install Modu from a link, or add it to .vscode/mcp.json. On first use VS Code opens a browser window to sign in to Modu (OAuth with automatic client registration).

Choose Install in VS Code below, or run MCP: Add Server from the Command Palette, pick HTTP (HTTP or Server-Sent Events) and paste the URL.
Or add the snippet to .vscode/mcp.json in your workspace. VS Code’s format uses a top-level servers object.
Sign in to Modu and allow access when VS Code asks, then use the Modu tools from Copilot Chat.

Install Server Setup in VS Code Install Tickets in VS Code

Good to know

On a Copilot Business or Copilot Enterprise plan, an organization admin must turn on the “MCP servers in Copilot” policy first. It is off by default.
The install links open VS Code, so they work on a computer with VS Code installed.

Checked 2026-10-23 against VS Code docs: MCP configuration, VS Code docs: MCP install URL, GitHub docs: MCP policy.

.vscode/mcp.json

{
  "servers": {
    "modu": {
      "type": "http",
      "url": "https://modu.ubits.tech/mcp"
    },
    "moduTickets": {
      "type": "http",
      "url": "https://modu.ubits.tech/mcp/tickets"
    }
  }
}

Terminal

code --add-mcp '{"name":"modu","type":"http","url":"https://modu.ubits.tech/mcp"}'

Windsurf (now Devin Desktop)

Windsurf became Devin Desktop on June 2, 2026. Its default agent, Devin Local, reads MCP servers from the Devin CLI config; the legacy Cascade agent reads mcp_config.json with a serverUrl field.

Devin Local: run devin mcp add -s user for each server, then devin mcp login modu to sign in to Modu. Devin Local asks before it calls any MCP tool.
Cascade: add the snippet to mcp_config.json (~/.config/devin/mcp_config.json on macOS and Linux, %APPDATA%\devin\mcp_config.json on Windows). Remote servers use serverUrl.
With a personal access token, add "headers": { "Authorization": "Bearer ${env:MODU_TOKEN}" } to the server.

Good to know

On Enterprise, an admin must turn MCP on in settings first.

Checked 2026-10-23 against Devin Desktop FAQ, Devin CLI: MCP configuration, Cascade MCP.

Devin Local (terminal)

devin mcp add -s user modu https://modu.ubits.tech/mcp
devin mcp add -s user modu-tickets https://modu.ubits.tech/mcp/tickets
devin mcp login modu

Cascade: mcp_config.json

{
  "mcpServers": {
    "modu": {
      "serverUrl": "https://modu.ubits.tech/mcp"
    },
    "modu-tickets": {
      "serverUrl": "https://modu.ubits.tech/mcp/tickets"
    }
  }
}

ChatGPT

ChatGPT on the web connects custom MCP servers as plugins, with read and write tools. It signs in with OAuth only, so personal access tokens can’t be used here.

Go to ChatGPT Plugins, choose the + button, then Add custom MCP server.
Enter Modu as the name. Under Connection, paste the Server Setup URL. Choose OAuth for authentication.
Review the risk warning, choose I understand and want to continue, then Create as a plugin. Install it and sign in to Modu.
In a chat, type @ and pick Modu. Repeat with the Tickets URL for the ticket tools.

Good to know

OpenAI’s help center lists custom MCP servers for ChatGPT Business, Enterprise and Edu on the web. In a Business workspace only admins and owners can create them. Check OpenAI’s help center for Plus and Pro.
Workspace permissions and Lockdown mode apply. ChatGPT asks you to confirm write actions by default.

Checked 2026-10-23 against OpenAI docs: custom MCP server, OpenAI Help Center.

Server Setup URL

https://modu.ubits.tech/mcp

Tickets URL

https://modu.ubits.tech/mcp/tickets

Grok and the xAI API

Grok reaches Modu three ways: a custom connector on grok.com, the Grok Build coding agent, and the xAI API’s remote MCP tool with a personal access token.

grok.com: go to grok.com/connectors, choose New Connector, then Custom. Enter the Modu URL and complete the sign-in. On Grok Business and Enterprise, a team admin provisions the connector first.
Grok Build: run grok mcp add --transport http with the Modu URL; it handles the sign-in.
xAI API: create a personal access token on /app/settings/ai, then pass Modu as an mcp tool to https://api.x.ai/v1/responses, sending the token in headers. Use allowed_tools to limit which Modu tools Grok may call.

Good to know

Grok Bot adds connectors from the plugins listed in Connect apps. Its docs describe no way to add your own MCP server URL, so Modu can’t be added to Grok Bot directly yet.
The xAI API calls Modu from xAI’s servers, and its require_approval option is not supported, so Grok calls the tools you allow without asking. Modu still holds destructive steps for approval in Modu.

Checked 2026-10-23 against xAI docs: remote MCP tools, Grok connectors, Grok Build: MCP servers, Grok Bot: connect an app.

Grok Build (terminal)

grok mcp add --transport http modu https://modu.ubits.tech/mcp
grok mcp add --transport http modu-tickets https://modu.ubits.tech/mcp/tickets

xAI API (curl)

curl https://api.x.ai/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -d '{
  "model": "grok-4.7",
  "input": [{"role": "user", "content": "Show my open tickets and summarise the oldest one"}],
  "tools": [
    {
      "type": "mcp",
      "server_url": "https://modu.ubits.tech/mcp/tickets",
      "server_label": "modu_tickets",
      "headers": {"Authorization": "Bearer '"$MODU_TOKEN"'"}
    }
  ]
}'

xAI API (Python, OpenAI SDK)

import os
from openai import OpenAI

client = OpenAI(api_key=os.getenv("XAI_API_KEY"), base_url="https://api.x.ai/v1")

response = client.responses.create(
    model="grok-4.7",
    input=[{"role": "user", "content": "List my Modu servers"}],
    tools=[
        {
            "type": "mcp",
            "server_url": "https://modu.ubits.tech/mcp",
            "server_label": "modu",
            "headers": {"Authorization": f"Bearer {os.getenv('MODU_TOKEN')}"},
        }
    ],
)
print(response.output_text)

Any other MCP client

Modu speaks the MCP Streamable HTTP transport (spec 2025-06-18), so any client that can add a remote MCP server can use it.

Add a remote (HTTP or Streamable HTTP) server with the Modu URL. Most clients use a JSON entry like the one below.
If the client supports MCP sign-in, it finds Modu’s sign-in from the server’s 401 response and opens it for you.
If it doesn’t, create a personal access token on /app/settings/ai and send it as Authorization: Bearer modu_pat_....

Good to know

Clients that only run local (stdio) MCP servers can’t connect to Modu directly.

Checked 2026-10-23 against MCP spec: transports, MCP spec: authorization.

Typical mcpServers entry

{
  "mcpServers": {
    "modu": {
      "type": "http",
      "url": "https://modu.ubits.tech/mcp"
    },
    "modu-tickets": {
      "type": "http",
      "url": "https://modu.ubits.tech/mcp/tickets"
    }
  }
}

With a personal access token

{
  "mcpServers": {
    "modu": {
      "type": "http",
      "url": "https://modu.ubits.tech/mcp",
      "headers": {
        "Authorization": "Bearer modu_pat_..."
      }
    }
  }
}

Try it

Things to ask

Once connected, ask in your own words. Your AI app picks the Modu tools it needs.

“Set up my new server as a developer community with roles and a ticket system”
“Show my open tickets in UBITS and reply to #12”
“List my Modu servers and give me an overview of the busiest one”
“Add a #releases channel that only the Maintainers role can post in”
“Summarise ticket #7 before I answer it”
“Claim ticket #9 and tell them we are looking into it”
“How is that last Modu run going? Undo it if it failed”

Safety

Your AI asks, Modu checks

The rules every connection follows

Modu plans every change
A setup request runs the same Modu agent as the chat. It inspects the server and plans each step with its risk level, and your AI app can read that plan before anything is approved.
Destructive steps are approved in Modu
Your AI app can approve a plan only when none of its steps is destructive. If one is, Modu returns a link to the run, and a person approves it inside Modu, never from the AI app.
Ticket text is untrusted
Ticket messages are written by your members, so Modu hands them to your AI app clearly marked as untrusted content, not instructions. Replies go out as Modu, labelled with the app that sent them, with mentions turned off.
Only your servers
A connection acts as you: it reaches only the servers you connected to Modu, or manage through team access, under the same plan rules as the app.
Revocable at any time
Disconnect an app or revoke a token on /app/settings/ai. Sign-in tokens last an hour and are refreshed while the connection stays allowed; personal tokens are shown once and stored hashed.
Credits work like chat
A setup request uses AI credits exactly as the same request in the Modu chat would. Your AI app's own usage is billed by that app, on its own plan.

Letters

Questions about connecting

What is the Modu MCP server?

MCP, the Model Context Protocol, is an open standard that lets AI apps use outside tools. Modu runs four MCP servers: Modu Server Setup (https://modu.ubits.tech/mcp) builds and changes your Discord servers, Modu Tickets (https://modu.ubits.tech/mcp/tickets) lists, reads, answers, claims and closes tickets, and Modu for Slack (https://modu.ubits.tech/mcp/slack) and Modu for WhatsApp (https://modu.ubits.tech/mcp/whatsapp) work in the Slack workspaces and WhatsApp numbers linked to your Modu plan.

Which AI apps can connect to Modu?

Any app that can add a remote MCP server over Streamable HTTP. This page has steps for Claude (web, desktop and Claude Code), OpenAI Codex, Cursor, VS Code with GitHub Copilot, Windsurf, ChatGPT and Grok through the xAI API.

Do I give my AI app my Discord password or token?

No. Your AI app opens a Modu sign-in page, you sign in with Discord as usual and approve the app on Modu's own consent page. The app receives a Modu token that you can revoke, never your Discord credentials. Apps without sign-in support can use a personal access token instead.

Can an AI app delete channels or roles through Modu?

Not on its own. Modu plans every change, and any plan with a destructive step can only be approved inside Modu. The AI app gets a link to the run, and a person approves it there.

Can Modu create a brand-new Discord server for me?

Discord stopped letting apps create servers on July 15, 2025. Ask your AI app to start a new server: Modu gives you the steps to create an empty server in Discord and the link to add Modu, and then sets it up from your request.

Does using Modu from an AI app cost credits?

Setup requests use AI credits exactly like the same request in the Modu chat, under the same plan rules. Your AI app's own usage is billed separately by that app.

How do I disconnect an AI app or revoke a token?

Open AI connections at https://modu.ubits.tech/app/settings/ai. It lists every connected app with a Disconnect button and every personal access token with Revoke. Either one stops access straight away.

Who can use the ticket tools?

You, for servers you connected to Modu, and teammates whose team access lets them manage that server. Replies are posted in the ticket channel as Modu, labelled with the AI app that sent them.

Destructive steps are approved in Modu, never from the AI app.

Read the documentation for how the agent plans and approves changes, or pricing for AI credits.

Connect your first AI app

Continue with Discord AI connections

Connect AI assistants to Modu

Modu’s MCP servers

Modu Server Setup

Modu Tickets

Modu for Slack

Modu for WhatsApp

How sign-in works

Connect with Modu

Personal access token

Set it up in your app

Claude (web and desktop)

Claude Code

OpenAI Codex

Cursor

VS Code with GitHub Copilot

Windsurf (now Devin Desktop)

ChatGPT

Grok and the xAI API

Any other MCP client

Things to ask

Your AI asks, Modu checks

The rules every connection follows

Questions about connecting

Connect your first AI app