{"version":"2.1.284","anchor":"waiting-for-pending-mcp-servers-per-response-budget-and-ex","canonical_anchor":"waiting-for-pending-mcp-servers-per-response-budget-and-ex","heading":"Waiting for slow MCP servers is now shared across one response","tier":"notice","area":"MCP","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.284\/e\/waiting-for-pending-mcp-servers-per-response-budget-and-ex","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.284","markdown":"### Waiting for slow MCP servers is now shared across one response\n\nSeveral tool calls in one response now share one wait for a connecting MCP server, and a server that already timed out can be skipped\n\n**Unclear.** When the skipping of already timed-out servers is switched on is not known.\n\n**What**\n\nWhen Claude calls a tool from an MCP server (an add-on that provides tools) that is still connecting, Claude Code waits for it for a limited time. That wait has changed:\n\n- All tool calls in one reply from Claude now share a single time budget. Before, each call could wait the full time on its own.\n\n- A server that used up a full wait without connecting is remembered, and later waits can skip it.\n\n- If Claude Code cannot work out which server a tool belongs to, it logs \"Not waiting for MCP tool\" and does not wait.\n\n**Why**\n\nA reply that makes many calls to a server that never connects no longer adds up to a long stall.\n\n- Area: MCP\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 1\/5"}