- Duplicate
- Answered
- Confirmed
- Yours
Short answer
How do you automate GitHub issue triage?
You connect GitHub once and ask. Your AI reads each open issue, searches your issues and your code for what it needs, and acts on it. No scripts, no cron jobs, no code to maintain. Every issue it reads lands in one of four places.
The payoff
Where every issue lands.
Each place below is a kind of done: what your AI already found and what it did about it.
Duplicate
Already reported here.
Your AI searches your open issues for the same report, same error, same endpoint, same symptom, and reads both before it says anything. When it finds the original it posts a pointer on the new issue, so the reporter lands in the right thread.
Answered
Your repo already says it.
How-do-I questions get answered on the issue itself, with the link to the place in your repo that already explains it: a README, a doc page, a closed thread.
Confirmed
The code confirms it.
A bug report gets read against the code it names. What your AI finds goes on the issue: the file, the cause, the related report. It does not fix it, because fixing is yours.
Yours
Needs a human.
What cannot be answered comes to you with the context attached, the file, the reproduction, the related issue, so you decide from the summary instead of digging.
The workflow
What your AI does when an issue arrives.
Triage is mostly reading and context, and that is exactly what your AI is good at. Four moves, repeated on every issue, while you do something else.
Your AI reads the issue
Point it at a repository and it lists the open issues, then opens each one and reads the whole report: the title, the body, and the comments already there.
list_repo_issuesget_issue_detailsIt checks the code
An issue that says the webhook fires twice is only words until someone reads the webhook. Your AI searches for the code the issue names and reads the file, so its answer is about your code, not a guess from the title.
search_github_codeget_file_contentsIt answers what it can answer
Duplicates, usage questions, known fixes: your AI posts the answer on the issue itself and links the original report or the documentation. It closes nothing, because closing is yours.
add_issue_commentIt brings you the rest
When an issue needs a human it comes to you with the context attached: the file, the reproduction, the related issue. And when the triage reveals a real bug, it opens the follow-up issue for you.
create_new_issue
Try it
Copy these to start.
Real prompts for issue triage. Copy one into the chat as soon as GitHub is connected, in your own words.
Starting points
These are opening lines, not scripts. Name your repository, your issue numbers and the outcome you want; your AI reads the same Connector either way, and every answer it gives links back to what it read.
GitHub Connector
You're all set. Choose your MCP client and follow the setup instructions.
https://edge.vinkius.com/vk_preview_fIryFqSn5ElX5r3SAhmNYdc5Yr0BXuXnnrfiP1oC/mcpClaude Desktop
Follow the steps below to connect in seconds.
- 1In Claude Desktop, open Settings → Connectors.
- 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
- 3Click Add and start a new chat — GitHub capabilities are ready to use.
{
"mcpServers": {
"github-mcp": {
"url": "https://edge.vinkius.com/vk_preview_fIryFqSn5ElX5r3SAhmNYdc5Yr0BXuXnnrfiP1oC/mcp"
}
}
}Triage the open issues in vinkius/api. Tell me which are duplicates, which you can answer, and which need me.
Read issue 482 in vinkius/api and then read the code it points at. Is the bug real?
Someone said the webhook fires twice on refunds. Search the payments code and tell me if you can reproduce it.
Answer the three oldest open issues in vinkius/api on the issues themselves, and bring me anything you can't answer.
Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline
Start here
Connect GitHub once, then ask.
One link in your AI's connector settings and GitHub is connected; your login stays encrypted on our side. The full walkthrough, every other capability, and the measured latency live on the Connector page. Come back here and paste a prompt.
Connect GitHub to your AIFAQ
How automated issue triage behaves.
- 01
How does my AI know an issue is a duplicate?
It searches your open issues for the same report, the same error, the same endpoint, the same symptom, and it reads both issues before it says anything. When it finds the original it posts a pointer on the new issue instead of a second answer.
- 02
Will my AI close issues without asking me?
No. Through this Connector your AI can read issues and post comments on them, and it closes nothing. When it cannot answer, it brings the issue to you with what it found.
- 03
Can it label or assign issues?
No. Through this Connector your AI reads your issues, posts answers on them, and opens a follow-up issue when triage finds a real bug. Labels, assignees and closing stay with you. The only thing it changes on an existing issue is the comment it posts.
- 04
What if the triage is wrong?
You see every response it posts, in the chat and in the activity log, and the Connector never deletes or closes anything. A wrong sort is a comment you correct, not a change you cannot undo.
- 05
Can it triage the backlog, or only new issues?
Both. Point it at today's new issues or the whole backlog; it works through them and reports what it did.
- 06
Which AIs can do this?
The ones you already use: Claude, ChatGPT, Gemini, and any client that takes a connector link. The same link works in all of them, so you switch AI without setting anything up again.
- 07
Does it work on private repositories?
Yes. You authorize GitHub once and your login stays encrypted on our side. Your AI sees only the connection link.
More questions about GitHub? The Connector page answers them. See everything the GitHub Connector can do
Connect GitHub to Claude, Cursor, ChatGPT & more
