--- title: "Validatar Agent Kits" slug: "validatar-agent-kits" updated: 2026-11-06T15:12:32Z published: 2026-11-06T15:12:32Z canonical: "docs.validatar.com/validatar-agent-kits" --- > ## Documentation Index > Fetch the complete documentation index at: https://docs.validatar.com/llms.txt > Use this file to discover all available pages before exploring further. # Validatar Agent Kits A Validatar Agent Kit is a folder you add to your AI coding tool so it can build Validatar tests the way an experienced Validatar engineer would. With a kit in place you can ask the agent to profile a table and suggest tests, turn a mapping document into a test plan, create standard and template tests, run them and explain the failures, package a test pack for another environment, or build a data source template for a source Validatar does not support out of the box. Kits are plain Markdown instructions and small Python scripts. They install nothing, need no package manager, and run on Python 3.9 or newer with its standard library only. ## Supported tools | Tool | Setup article | Download | | --- | --- | --- | | Claude Code | [Validatar Agent Kit for Claude Code](/docs/validatar-agent-kit-claude-code) | [Marketplace](https://admin.cloud.validatar.com/marketplace/95) | | OpenAI Codex (CLI, IDE extension, app) | [Validatar Agent Kit for OpenAI Codex](/docs/validatar-agent-kit-codex) | [Marketplace](https://admin.cloud.validatar.com/marketplace/96) | | GitHub Copilot (VS Code agent mode, Copilot CLI, cloud agent) | [Validatar Agent Kit for GitHub Copilot](/docs/validatar-agent-kit-github-copilot) | [Marketplace](https://admin.cloud.validatar.com/marketplace/97) | | Cursor (editor and CLI) | [Validatar Agent Kit for Cursor](/docs/validatar-agent-kit-cursor) | [Marketplace](https://admin.cloud.validatar.com/marketplace/98) | | Snowflake Cortex Code (CLI and Desktop) | [Validatar Agent Kit for Snowflake Cortex Code](/docs/validatar-agent-kit-cortex-code) | [Marketplace](https://admin.cloud.validatar.com/marketplace/99) | | Google Antigravity (and Gemini CLI) | [Validatar Agent Kit for Google Antigravity](/docs/validatar-agent-kit-antigravity) | [Marketplace](https://admin.cloud.validatar.com/marketplace/100) | | Kiro (IDE and CLI) | [Validatar Agent Kit for Kiro](/docs/validatar-agent-kit-kiro) | [Marketplace](https://admin.cloud.validatar.com/marketplace/101) | | Devin (Desktop and CLI) | [Validatar Agent Kit for Devin](/docs/validatar-agent-kit-devin) | [Marketplace](https://admin.cloud.validatar.com/marketplace/102) | Each kit is a free download from the Validatar Marketplace. ## Two access modes Every kit works in two modes. Which one applies depends on the tokens you give it. The first time you connect, the agent checks which tokens are set; if there is no API token, it explains what one adds and asks whether you can create one. It records the mode in the kit's working folder and follows the matching steps from then on. | | MCP only | MCP + API | | --- | --- | --- | | Talks to Validatar through | The [Validatar MCP server](/docs/connecting-to-the-validatar-mcp-server) | The MCP server, plus the public REST API through the kit's Python scripts | | Tokens you create | One MCP token | One MCP token and one API token | | Explore the catalog, query data, create standard and template tests, run tests, set custom field values | The agent does it | The agent does it | | Update an existing test, materialize template tests, run a folder or job, download failed rows | The agent tells you which screen to use | The agent does it | | Export and import test packs, refresh catalog metadata, import data source templates, create custom field definitions | The agent tells you which screen to use | The agent does it | | Template tests selected by metadata filters | You switch the selection in the UI | The agent creates them directly | Start with MCP only if your organization limits API tokens. Add an API token later and the same kit does more of the work: set the variable, restart the tool, and ask the agent to reconnect. ## What a kit contains | Part | What it does | | --- | --- | | Always-on instructions | How to reach your instance, the access modes, the safety rules, and which skill handles which request | | Ten skills | Connect, explore, test plan, standard test, template test, bulk tests, run and review, test pack, data source template, custom fields | | Reference files | Test design patterns, template test Handlebars, test pack and data source template formats, custom fields, known platform behaviors, and a reference of every MCP tool and REST endpoint | | Python scripts | Setup check, MCP calls and previewed batches, configuration checks, test pack inspection and cleanup, data source template scaffolding and checks, and the REST steps | | Worked examples | Standard and template test configurations, each run on a Validatar instance before release | The agent keeps its work in a `validatar-work/` folder: test plans, the exact configuration it sent for every object, batch files, exports, and `changes.jsonl`, a log of every change it made. You can review, version and rerun all of it. ## How the agent works with your instance 1. It names the instance and project before its first change and waits for your confirmation. 2. It shows each test design (SQL, expected result, severity, quality dimension, folder) before creating it, and runs the query first to predict the result. 3. It creates the test, runs it, and compares the result with its prediction. 4. Every change made through MCP can be reverted, and every change the scripts make is logged. Queries the agent runs against your data sources are read-only by instruction. The Validatar user behind the token is the real boundary, so use a user whose access matches the work: a sandbox project on a non-production instance is the right place to start. ## Tokens and security - Tokens are read from environment variables (`VALIDATAR_MCP_TOKEN`, and `VALIDATAR_API_TOKEN` for MCP + API). No kit file holds a token, and the tool configurations reference the variable rather than the value. - Create tokens in Validatar under your user profile, **User Tokens**. MCP tokens have type MCP; API tokens have type API and carry scopes (Authoring, Execution, Data Steward, Catalog). - Changes made through MCP are recorded as AI Actions that administrators can review. - The `validatar-work/exports/` folder can hold rows from your data (failed-row downloads, exported tests). The kit's `.gitignore` keeps it out of version control. ## Tools that cannot connect yet Cortex Code in Snowsight and Databricks Genie Code accept only OAuth or bearer-token MCP servers, and the Validatar MCP server authenticates with its own header. Use the Cortex Code CLI or Desktop app instead of Snowsight. ## Updating a kit Kits carry a version and the Validatar version they were verified against (see `validatar/KIT.json`). New kits follow Validatar releases that add MCP tools or API capabilities. To update, download the new kit and replace the instructions file, the `validatar-*` skill folders and the `validatar/` folder. Keep `validatar-work/`. ## Related - [Connecting to the Validatar MCP Server](/docs/connecting-to-the-validatar-mcp-server) - [MCP Tools Reference](/docs/mcp-tools-reference)