OpenAI Agents SDK v0.23.0 ships MCP listing controls
SDK Release Adds Developer Control Over MCP Discovery
OpenAI released Agents SDK v0.23.0 on October 1, 2026, introducing a new configuration parameter that gives developers fine-grained control over how agents discover and list tools via the Model Context Protocol (MCP). The release notes specify: "mcp: add configurable MCP listing page limits" GitHub.
The feature centers on a `max_list_pages` parameter, defined as a positive-integer limit on successful pages returned during automatic `list_tools()` or `list_prompts()` calls GitHub. This addition addresses a practical constraint in agent workflows: uncontrolled tool enumeration can consume bandwidth, latency, and compute resources when MCP servers expose large tool catalogs.
Why Listing Limits Matter for Agent Operations
As agents increasingly act autonomously in production environments, the ability to bound discovery operations becomes critical. An agent connecting to a corporate MCP server with thousands of available tools doesn't need to enumerate all of them to function effectively. The new parameter allows developers to cap the pagination depth—ensuring agents retrieve only the tool subset necessary for their task, reducing unnecessary round-trips and improving responsiveness.
The release reflects a broader pattern in the Agents SDK: moving from all-or-nothing capability exposure toward granular, operator-configurable constraints. Earlier versions shipped with fixed defaults that sometimes proved too permissive or too restrictive depending on deployment context.
Implementation and Usability
Developers using the Python SDK can now pass `max_list_pages` when initializing MCP clients or within agent configuration blocks. The parameter is backward-compatible—omitting it preserves existing behavior, so existing agents continue to work without modification GitHub.
The change is minimal in scope but signals OpenAI's incremental approach to the Agents SDK: each release adds a small number of concrete, developer-requested features rather than broad capability claims. This aligns with the editorial focus of shipping, not roadmap.
Broader SDK Trajectory
v0.23.0 continues the SDK's release cadence established earlier in 2026. The tool remains under active development, with bug fixes and minor enhancements prioritized based on production usage patterns. Developers building multi-agent systems or integrating agents into existing MCP ecosystems should review the release notes to determine whether tighter listing controls improve their performance profile.
The SDK is available via pip and the official OpenAI Python package repositories.