{"version":"2.1.251","anchor":"eval-sandboxes-run-with-claude-ai-mcp-servers-off","canonical_anchor":"eval-sandboxes-run-with-claude-ai-mcp-servers-off","heading":"Eval sandboxes run with claude.ai MCP servers off","tier":"internal","area":"Plugin Eval","scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.251\/e\/eval-sandboxes-run-with-claude-ai-mcp-servers-off","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.251","markdown":"### Eval sandboxes run with claude.ai MCP servers off\n\nEval sandboxes force claude.ai MCP servers off so they can't reach real servers.\n\n**What**\n\nThe internal evaluation harness now forces `ENABLE_CLAUDEAI_MCP_SERVERS` to false and pins its own temp directories, so evals cannot reach real MCP servers or share scratch space with the host.\n\n**Details**\n\n- Pins `TMPDIR`, `TMP`, `TEMP` and `CLAUDE_CODE_TMPDIR`.\n- Affects the eval harness only; it is not a runtime flag for normal use.\n\n**Evidence**\n\n`ENABLE_CLAUDEAI_MCP_SERVERS: \"false\"`\n\n- Area: Plugin Eval\n- Names: `ENABLE_CLAUDEAI_MCP_SERVERS`\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 3\/5"}