Connect Vindex to GitHub Copilot
GitHub Copilot works in VS Code and as a coding agent on your repository. Connected to Vindex, either can read your rules and check its output against them.
Three steps, about 4 minutes. Open to every registered Vindex user; new users create an account during sign-in, and a new account includes 100 free critique credits.
The walkthrough
Add the Vindex MCP server
In VS Code, open the Command Palette and run MCP: Add Server. Choose HTTP, paste the server URL, name it vindex, and pick Workspace to write .vscode/mcp.json for the repository, or Global to use it everywhere.
For the Copilot coding agent, open the repository's Settings, then Copilot, then Coding agent, and add the same server under MCP configuration so the agent can call Vindex while it works an issue.
{
"servers": {
"vindex": {
"type": "http",
"url": "https://mcp.vindexgateway.com/mcp"
}
}
}Sign in with your Vindex account
Start the server from the MCP view or the inline Start action in mcp.json. VS Code asks to allow the sign-in and opens your browser on the Vindex sign-in; use the account you use for the Vindex Portal, or create one to start on the free plan.
When the browser confirms, the server shows as running with its tools. In Copilot Chat, open the tools picker in agent mode and make sure the Vindex tools are selected.
Run your first critique
Ask your agent to list your rules, then to check a draft against one. Reading your rules is free, and each critique uses one critique credit.
The result comes back in the chat: pass or fail for each rule, with the failing passage quoted.
List my Vindex rules.Check this draft with Vindex against the "Email tone" rule: <paste the draft>What comes back
Each rule gets a pass or fail, and a failure quotes the passage that broke the rule. A rule set also returns one overall result and the names of the rules that failed. The result carries the attempt number and your credits left.
GitHub Copilot does not block anything on its own. It gets the result and decides what to do next: usually, rewrite the quoted passage and run the critique again. Critique results and the loop has every field.
Tell the agent to run the loop
Copilot Chat and the coding agent both read .github/copilot-instructions.md. Put the standing instruction there.
Before you send or save any customer-facing text, check it with the
Vindex validate tool against the rule set named below.
Rule set: <your rule set or rule name>
If a rule fails, rewrite only the passage the result quotes, then run
validate again. Stop when every rule passes, when the attempt limit is
reached, or when credits run out. Then tell the user which rule, if any,
is still failing. Never describe an output as checked unless validate
returned a result for it.An agent that follows it runs the loop on its own, and the server refuses a call over the attempt limit. The standing instructions guide has more.
Troubleshooting
The server shows as running but Copilot does not use it
Open the tools picker in Copilot Chat agent mode and select the Vindex tools. Ask mode does not call tools; switch to agent mode.
The coding agent cannot reach Vindex
Confirm the server is in the repository's Copilot coding agent MCP configuration and that the sign-in for it completed. The agent uses the repository configuration, not your local mcp.json.
The agent answered without running a critique
Ask for it in the request: say "check it with Vindex". Put the standing instruction in your client's instructions file so it applies on every task. An agent chooses whether to call a tool, so treat an output as checked only when a result came back.
A critique came back as one sentence with no pass or fail
That sentence is a refusal or an error, such as no credits left, a rule that is not active, or a timeout. Ask your agent to check your credit balance, or try again in a moment.
Using a different client?
Scoring user stories too?
The same connection and the same sign-in also score a story against INVEST. That is part of Integrated Vindex.
Stories in your AI client