ChatGPT Agent

Verification notes: OpenAI adopted MCP (originally open-sourced by Anthropic, Nov 2024) across its platform in March 2025 — confirmed via official docs covering the Responses API, Agents SDK, ChatGPT Developer Mode, Apps SDK, and Codex. Re-checked this pass: MCP support has expanded rather than changed in kind — OpenAI's own docs now cover Apps & Connectors (added Sept 2025), native Responses API connections to remote MCP servers, and a 'developer mode' for building and publishing MCP connectors that requires OAuth 2.1 with Dynamic Client Registration (bearer tokens alone are rejected). Still no A2A support confirmed. Note on audit status: OpenAI's own Trust Portal lists SOC 2 Type 2, ISO/IEC 27001 family, and ISO/IEC 42001 certifications, but these cover the API Platform and ChatGPT Enterprise/Edu/Team broadly rather than naming the agentic ChatGPT Agent feature specifically, so audited stays false here rather than being inferred from org-wide certification — see Copilot Studio for a case where a certification's scope was confirmed to explicitly include the agent product before that flag was flipped. Re-checked this pass: OpenAI's own docs confirm the connector_id parameter is deprecated for models released after September 1, 2026, replaced by server_url (remote MCP) or tunnel_id (local MCP via the new Secure MCP Tunnel) — an implementation-level change to how MCP connections are wired up, not a change to whether MCP is supported. OpenAI's SOC2 Type II period now runs July 2025–June 2026 per its Trust Portal, still organization-wide rather than naming ChatGPT Agent specifically, so audited stays false. No flag changes.