On Rithub, label an issue agent and it opens a pull request that closes it — a built-in issue-to-PR loop, no bot account or Action to configure. Prefer your own agent? Point Claude Code, Cursor, or Codex at Rithub's native MCP endpoint and it can do the same thing by hand, scoped to exactly one repo.
Turning an issue into a working pull request is a real, common workflow now — GitHub's own Copilot coding agent does it when you assign an issue to it, and several open-source projects (orchestrator agents, webhook-triggered coding bots) exist to do the same thing against any repo. All of them need something configured first: a coding-agent subscription, a bot account, a webhook, or a CI job that watches for a label.
| Step | What happens |
|---|---|
| 1 | Open an issue, describe the fix |
| 2 | Add the agent label (once per repo: turn the feature on in repo settings) |
| 3 | Rithub opens a pull request against the issue |
| 4 | Merging the PR closes the issue, same as any normal pull request |
If you'd rather keep a specific coding agent in the loop end to end, mint a repo-scoped token at rithub.dev/agent-tokens (pick pr scope: branches, commits, pull requests, issues, comments — no merge, no settings) and point your agent's MCP client at https://rithub.dev/mcp:
claude mcp add --transport http rithub https://rithub.dev/mcp --header "Authorization: Bearer <token>"
From there the agent can read the issue, branch, commit, and open the PR itself — the same loop, with whichever coding agent you already use.
This isn't a bigger agent-tooling ecosystem than GitHub's — Copilot's coding agent and the third-party orchestrators have more surface area today. What's real on Rithub: the issue-to-PR loop is a repo setting, not an integration project, and the MCP endpoint is live and scope-fenced, not a demo.
Mint a scoped agent token Import a repo from GitHub