Score the story where you write it. Before it becomes anyone’s work.
The story your coding agent just built from was never checked. The one you drafted in chat looked fine until refinement. The Vindex MCP server puts the check inside the assistant: one URL, one sign-in, and Claude, ChatGPT, Codex, Cursor, or Copilot can ask Vindex whether a story is ready, and hear why not, before anything is built on it.
Open to every registered Vindex user. Nothing to install, no key to paste. Connecting takes about two minutes; the free plan covers up to 10 stories a month.
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%
The story was written in a chat window. The defect was written there too.
More stories start in an assistant than in a tracker now. A product owner drafts in Claude or ChatGPT and pastes into Jira. A coding agent in Cursor or Codex is handed a story and starts building. In both cases the story is read by a machine that will do exactly what it says, and nobody has asked whether what it says is enough.
An agent does not stop at a vague story. It fills the gaps with plausible guesses and ships them, and the defect report lands weeks later with a root cause that points straight back to the ticket. A product owner who drafts in chat gets the same failure more slowly: the story looks finished, reaches refinement, and falls apart in front of the team.
The Vindex MCP server puts the check in the same window. The assistant hands Vindex the story as written and gets back a finding, a percentage, a band, and the reasons. Vindex judges; what you and your assistant do with the judgment is yours. It does not rewrite the story and it does not block anything.
Two minutes to connect. Then it is one sentence away.
The Vindex MCP (Model Context Protocol) server is a tool your assistant can call. One story per call, by design: the assistant is the right place to fix one story, and the plugins and the Vindex Portal are the right place to see a backlog.
- Step 1
Add one URL to your assistant.
Paste the Vindex MCP server address into Claude, ChatGPT, Codex, Cursor, or GitHub Copilot. Nothing to install, no key to manage.
- Step 2
Sign in with your Vindex account.
Your browser opens once. Use the account you use for the Vindex Portal, or create one and start on the free plan. Your organization is your company.
- Step 3
Ask for a score.
“Score this story with Vindex.” The result reads finding first, then the percentage and band, then the six dimensions if you want the reasons.
Any MCP client. These have their own setup pages.
The same three steps everywhere: add the server, sign in, score a story. Each page uses the platform’s own words for where the settings live and where the result appears.
Claude Code
Coding agentOne command in the terminal, then a ready check on every story your agent picks up.
Connect ClaudeClaude Desktop and claude.ai
AssistantA custom connector in Settings; score any story you draft in the chat.
Connect ClaudeChatGPT
AssistantA connector in ChatGPT settings; check the story before you paste it into Jira.
Connect ChatGPTCodex
Coding agentRegister the server from the Codex CLI and give the agent a rule in AGENTS.md.
Connect CodexCursor
Assistant + agentOne entry in mcp.json; Agent checks the story before it touches the codebase.
Connect CursorGitHub Copilot
Assistant + agentAn MCP server in VS Code for Copilot Chat, and in repository settings for the coding agent.
Connect GitHub
Using something else? Any client that speaks MCP over HTTP works.
Everything your client needs is one address.
The server is a thin adapter over the same scoring engine every Vindex integration uses. It holds no scores and no keys; it checks who you are and asks Vindex.
Read the full referencehttps://mcp.vindexgateway.com/mcpMCP over Streamable HTTP. This is the only value you paste into a client.
OAuth 2.1 through the same sign-in as the Vindex Portal. Your browser opens once; the client keeps the session. Open to every registered Vindex user, and new users create their account on first sign-in. No API key exists to copy or leak.
score_storyOne story, scored against INVEST. Returns the percentage, the band, six dimensions with reasons and suggestions.get_usageYour company’s stories this month against its allowance. Re-scores never count.explain_investThe six dimensions and the three bands, explained. Uses no allowance.
Guides for assistants and agents
How to let an agent check a story before it builds, how to check a draft before it reaches the tracker, and how to read what comes back.
How do I let my coding agent check a story is ready before it builds?
Give Claude Code, Codex, Cursor, or Copilot the Vindex MCP server and one rule: score the story before writing code, proceed on Healthy, report and ask otherwise. What the agent sends, what comes back, and a real story it would have stopped on.
How do I put a Vindex ready check into my agent's standing instructions?
The three-line rule that makes the ready check permanent, and the file each agent reads it from: CLAUDE.md, AGENTS.md, Cursor rules, or copilot-instructions.md. How to make the threshold your team's and how to confirm the rule fires.
How do I check a story before it goes into Jira, Azure DevOps, or Asana?
Draft the story in your assistant, run the score-story command, and fix what the result names before the story exists in the tracker. Nothing downstream, no refinement slot, no plugin comment, ever sees the version that was not ready.
Check the next story before anyone builds on it.
Two minutes to connect. One sentence to score. Free re-checks every time you fix what it found.