vinktar.ai

What you can ask

• Find the biggest drop in a conversion funnel and what distinguishes the people who drop from the ones who convert.Copy

  Example path query_funnel→get_alert

• Set Vinktar up in your app: the SDK, the tracking plan, the check, and where each environment variable goes. With your code open it does the work; in a chat it gives you the lines to add.
• Root-cause one error issue: what breaks, for whom, since when, what it costs, and a proposed fix.
• A short tour of the sample project Vinktar generated: what the made-up app does, its worst error, its funnel, and what to ask next.
• What happened in a project this week: traffic, key events, the team's dashboards, and the errors worth fixing, with links.
• New, returning and rising errors in the last 7 days, the worst ones opened with their stack trace and what the user did before.
• Brief, confirm, place and narrate a call to a business, then report what was confirmed.
• Read back a finished call: the outcome, what was confirmed, the answers, the cost.

MCP connector

Server URL

https://mcp.vinktar.com/mcpCopy

Login

API key

Protocol

MCP 2025-11-25

Instructions

Vinktar holds a product's analytics events and its error tracking (issues, stack traces). Use it to answer questions about how the product is used and what is breaking.

Until the person signs in, the tools are listed but every call except get_install_guide, start_project and get_start_status asks for sign-in, which only the person can do. When setup should not wait for that, start_project makes a project nobody owns yet: a write key to set the app up with now, and a claim link the person opens later to keep it with its data. If a call fails for sign-in, tell them where to connect Vinktar in the app you are running in, then carry on:

Claude on the web, Claude Desktop (chat, Cowork and the Code tab) and Claude Code on the web: open https://claude.ai/customize/connectors, find Vinktar, choose Connect, sign in and approve. A chat or session that was already open may not pick it up; if the tools still ask for sign-in, start a new one.

Claude Code in a terminal or an editor: run /mcp, pick vinktar, choose Authenticate.

Codex: run codex mcp login vinktar.

Cursor asks on its own when a tool is first called.

In the browser they sign in to Vinktar or create an account on the spot, leave Build and Configure switched on, and approve.

Setting a project up: call get_install_guide and follow it (also the prompt install_vinktar and the resource vinktar://guide/install). In a chat with no codebase open the guide still applies: it starts with what to do there, so call it instead of saying setup needs a coding agent. Call get_project_keys for the project's key instead of asking the user to paste one, define each event you add with define_event in the same step (names and property keys in snake_case, object then past-tense action, like report_exported; a one-sentence description; get_install_guide with section: "conventions" has every convention), write what the product is with set_project_notes, and call get_setup_status to check what actually arrived; repeat it after each fix. Setup calls (get_install_guide, list_projects, get_project_keys, get_setup_status, get_schema, define_event, set_project_notes) never count against the workspace's agent allowance. Once signed in, no tool creates projects or keys: a person does that in Vinktar.

Show all

Tools

Skills

1 in the Claude listing

• setupSet up Vinktar product analytics and error tracking in this codebase end to end, from picking the SDK to checking the data arrives and saying where each environment variable goes, and connect Vinktar