memories/ai-operations-mcp-connector-setup-6e6565bb.md

memory

Purpose

Provide a short reference for connecting an AI platform to an approved external data source through MCP. This acts as an auxiliary guide to the authoritative [[AIPM Playbook: MCP Connector Setup & Integration Configuration]].

Use Cases

Use this when a requester needs an AI assistant to connect to a named external system, access live data, or activate an MCP server in [[Glean]], [[Claude]], or [[ChatGPT]].

Setup Flow

  • Identify the exact source system, endpoint, intended users, and business use case.
  • Confirm whether the system already has a native integration before introducing an MCP connector.
  • Confirm the required admin credentials, API access, client configuration, and service-account involvement.
  • Activate the connector separately in each relevant AI platform.
  • Restrict access to the requester or an approved group where platform capabilities allow it.
  • Remove or replace outdated connector versions when migrating.
  • Test authentication, authorization, representative reads, and expected failure behavior end to end.
  • Ask the requester to confirm successful use before closing the request.

Escalation and Decision Points

  • If the connector requires a new custom action, automation, or material implementation, promote the work to APMT as a CB or CBX item.
  • If the issue is caused by credentials, permissions, or an agent-version mismatch, resolve or route it through the relevant owner.
  • Involve Security, Legal / Compliance, or Corporate IT when the connector exposes sensitive data or requires enterprise credentials.

Guardrails

  • Scope access to the smallest practical audience and data set.
  • Do not put credentials or secrets in tickets, prompts, repositories, or shared documentation.
  • Keep the requestor-facing AIPM record linked to any APMT delivery item.
  • Document the connected system, platform, owner, and validation result without recording secret values.