Skip to content
Console

Credentials for agents

Follow Create an app and a credential and create a separate application or credential for each agent, so it can be revoked on its own.

Start read-only and add scopes as the agent’s tasks need them.

Agent task Scopes Data access flags
Explore what’s available website_read none
Investigate conversations and transcripts website_read, conversation_read, message_content_read, contact_read_pii message_content, pii
Look up contacts website_read, contact_read, plus contact_read_pii for names and emails pii for personal fields
Report on AI-handled conversations website_read, reports_read, conversation_read none, or message_content and pii for content-based measures
Review AI feature configuration website_read, config_read none
Follow change events website_read, event_read none

Contacts, configuration, AI features and event tools need an unrestricted application, with no channel, team or subsite restrictions.

Option How it works Trade-off
Installation token, recommended to start The agent sends the credential’s igw_live_... token on every request. Simplest setup. The token lasts until the credential expires, so use a short expiry.
Short-lived OAuth token A helper exchanges the credential for a 15-minute token bound to /mcp, and fetches a new one when it expires. Tokens sent to the server are short-lived, work only at MCP, and can be narrowed to fewer scopes. The client secret still has to be on the machine.

Store the token in an environment variable and reference it from the client configuration:

Client Reference
Codex bearer_token_env_var = "INSIDE_GATEWAY_TOKEN"
Claude Code "Authorization": "Bearer ${INSIDE_GATEWAY_TOKEN}"
Cursor "Authorization": "Bearer ${env:INSIDE_GATEWAY_TOKEN}"

Config files that use these references are safe to commit.