Skip to main content
Halo works as a connector for Claude: add it by URL today, or from the Connectors directory once Halo is published there. Once connected, Claude can read and answer your support conversations, look up any contact or company and replay their history, review negative feedback and unresolved questions, draft and launch outreach, and work your customer-success pipelines. Every action that changes a record, sends a message or launches an outreach is announced before it runs, and outreach activation requires the previewed recipient count.

​
Before you connect

  • A Halo workspace. Start one if you do not have one.
  • To grant write access (replying to customers, changing records, launching outreach), you need the matching permissions in Halo yourself, for example inbox: reply to reply to tickets or automation: send to activate a campaign. The consent screen only offers what you hold, and every write re-checks it.

​
Connect from the Connectors directory

Once Halo is published in the directory:
  1. In Claude, open Customize > Connectors and search for Halo.
  2. Choose Connect. Claude opens the Halo consent screen in a new window.
  3. Sign in to Halo if asked, then choose the permissions Claude should have. Read permissions are pre-selected; write permissions start unchecked.
  4. Choose Authorize. Claude returns to the conversation with Halo connected.
Tokens refresh automatically and expire after an hour. To revoke access, remove the connector in Claude; the refresh token is discarded with it. A Halo admin can also remove the operator who connected it from the workspace, which invalidates every token they granted.

​
Connect as a custom connector

If your organization installs connectors by URL, use:
In Claude: Customize > Connectors > Add custom connector, paste the URL, and follow the same consent flow. No client ID or secret is needed: Halo supports Dynamic Client Registration and Client ID Metadata Documents, so Claude registers itself.

​
Connect from Claude Code

Run /mcp inside Claude Code to complete sign-in and check the connection status. Claude Code is where the support loop closes: pull the tickets customers opened about a bug, fix it in the codebase, and reply to each conversation from the fix.

​
What you can ask

“Show me the open tickets about invoices and draft a reply to each.”
“Who at Acme has written in this quarter, and what did we tell them?”
“What went badly in support this week, and what should we fix first?”
“Write a renewal reminder series for accounts renewing next month and send me a test.”
“Which onboardings are stuck? Log a note on each and create a follow-up task.”
“Get the install snippet and add the Halo widget to our Next.js layout.”

​
What the connector can do

Every tool is annotated so Claude knows which ones only read (it runs them immediately) and which ones change state or send messages (it asks first). Read the full list under Scopes & Permissions.

​
Permissions

The consent screen lists every permission Claude asks for, grouped into Read and Write. Write permissions are never implied by read ones, and the connector can never do more than the person who connected it: each write re-checks that you hold the matching permission in Halo. Integration secrets never leave Halo; Claude only ever sees tool results. See Scopes & Permissions for the scope list and token lifetimes.

​
Troubleshooting

  • A write tool says a permission is missing: disconnect and reconnect the connector, then tick the write permission on the consent screen. Write permissions start unchecked, and only operators who hold the matching Halo permission can grant them.
  • Claude says the connector does not require sign-in: you pasted a different URL. The connector URL is https://app.haloagents.ai/api/mcp; the /chatgpt surface is for the ChatGPT plugin.
  • No tools appear: your workspace’s plan does not include that feature, or the MCP server was disabled for your workspace. Ask your Halo admin.
  • Still stuck? Email support@haloagents.ai.