Your assistant asks. Vindex answers in the same call.
The Vindex MCP (Model Context Protocol) server is a tool your assistant or coding agent can call. It sends one story as written and gets back the finding, the percentage, the band, and the reasons, before the story reaches a tracker or an agent builds on it.
- Step 1
Your client connects to the server
You add one URL to Claude, ChatGPT, Codex, Cursor, or GitHub Copilot. The client discovers three tools (score_story, get_usage, explain_invest), two reference resources, and one prompt. Nothing is installed on your machine and no key is stored.
Pick your client - Step 2
You sign in once
The client opens your browser on the Vindex sign-in, the same one the Vindex Portal uses. Your organization is your company, so the allowance you see is your team's. New users create an account on this screen and start on the free plan.
- Step 3
The assistant sends the story as written
When you ask for a score, the assistant splits the story into title, description, and acceptance criteria without changing a word, adds the tracker key if you gave one, and calls score_story. Vindex scores it through the same engine every integration uses and answers in the same call, usually in two to ten seconds.
- Step 4
The result reads finding first
The assistant shows the plain-language finding, then the overall percentage and its band (Healthy, Needs review, At risk), then the six dimensions with reasons and suggestions if you ask. The band is Vindex's suggestion. What happens next is up to you.
What the exchange looks like.
A real story from the Northwind sample backlog, scored through the live server. The finding comes first because it is the one change that moves the score most; the numbers back it up.
You
Score NW-108 with Vindex: “Warehouse scanning should be faster and better. The scanning process in the warehouse is slow and the team complains about it. Make scanning faster and improve the experience for the warehouse staff.”
Assistant
NW-108: 38% (At risk)
Add measurable acceptance criteria with baseline and target metrics (e.g., p95 scan-to-confirm time, scans/hour, error rate) and specify devices/network conditions for measurement.
- Independent40%
- Negotiable60%
- Valuable60%
- Estimable20%
- Small20%
- Testable30%
One command for whatever is in the chat.
The server publishes a score-story prompt. Clients surface it as a slash command or a quick action. It tells the assistant to take the story exactly as it stands in the conversation, whether you pasted it, the assistant drafted it, or you wrote it together, split it into title, description, and acceptance criteria without changing a word, and score it.
The prompt contains no instruction to rewrite or improve. Vindex judges the story as written. If you edit it and ask again, that is your workflow, and the re-check is free.
Counted once. Re-checked free.
A story counts against your company’s monthly allowance once. Every later score of that story is free, in any month, so the loop of fix, re-check, fix again costs nothing. Pass the tracker key when you have one (the Jira key, the Azure DevOps work item id, the Asana task) and the assistant and the tracker plugin agree on which story they are talking about.
Ask your assistant to check usage and Vindex reports stories counted this month, the allowance, and what remains. The free plan covers up to 10 stories a month; the arithmetic is the same the Vindex Portal shows.
What Vindex never does from your assistant.
- Vindex does not rewrite the story. It returns reasons and suggestions; the assistant and you decide what to change.
- Vindex does not write anything back to Jira, Azure DevOps, or Asana from the assistant. The tracker plugins own write-back.
- There is no batch tool. Ten stories are ten calls, presented one at a time, so each result is readable and no call outruns the client's timeout.
- Nothing blocks. A low band is a suggestion; if a coding agent stops on At risk, that is a rule your team wrote into its instructions.
Story text is sent to Vindex to be scored and for nothing else. Vindex does not train models on customer data. Read more on the security page.
Connect the assistant you already use.
Pick your client, paste one URL, sign in, and score your first story. About two minutes.