GitHub MCP Server Can Now Write Fully Triaged Issues — and Flag Duplicates Before They're Filed

TL;DR: The GitHub MCP server (June 18) can now read and write issue fields — AI agents create fully triaged issues with priority, area, and sprint set at filing time, and detect duplicates before submitting.

If you use Claude Code with the GitHub MCP server to file issues from test failures, code review findings, or agent workflows, two June 18 updates close the last manual step:

Issue fields read/write. Agents can now set custom project fields at creation — priority, area, due date, sprint, whatever fields your GitHub Projects have configured. A git bisect agent that finds a regression can file the issue fully triaged; no human follow-up required. Agents can also filter existing issues by field values, useful for "is there already a high-priority auth issue open?" checks inside a workflow.

Duplicate detection (public preview). Before filing, the agent can check for near-match open issues. On busy repos this prevents a backlog of "already reported" duplicates from agent-generated issue streams.

Both features are behind the remote_mcp_issue_fields flag, auto-enabled in Insiders Mode. To enable manually, add remote_mcp_issue_fields: true to your MCP server config.

Why it matters: Issue creation was the last unstructured step in automated developer workflows — agents that could write code, run CI, and open PRs couldn't file clean actionable issues without a human triage pass. That gap is now closed.

Sources: GitHub changelog — issue fields MCP support (June 18), GitHub MCP server