# AI Resources
Canonical page: https://www.adaptabletools.com/docs/getting-started-ai
- AdapTable uses AI to help you create and manage applications faster, and to a higher standard
- Four Agent Skills do that work — set up, develop, upgrade and debug — each from a single command to your agent
- AdapTable does not supply an agent - the skills work with the agent you already use
- They work as soon as they are installed, and they drastically cut the time that work takes
- The Skills follow the AdapTable version installed in your project
AdapTable (since 24) provides a set of powerful AI Resources.
These allow you to create, develop, debug and upgrade your applications faster and better than ever before.
The main elemment is a set of four [Agent Skills](https://www.adaptabletools.com/docs/getting-started-ai-skills/index.md) which cover the essential work done by a developer.
You give your agent one command, and the Skill carries it out:
- [adaptable-setup](https://www.adaptabletools.com/docs/getting-started-ai-skills-setup/index.md) — add AdapTable to an application
- [adaptable-dev](https://www.adaptabletools.com/docs/getting-started-ai-skills-dev/index.md) — build and change AdapTable
- [adaptable-upgrade](https://www.adaptabletools.com/docs/getting-started-ai-skills-upgrade/index.md) — move to a newer AdapTable
- [adaptable-debug](https://www.adaptabletools.com/docs/getting-started-ai-skills-debug/index.md) — find out why a running Grid is wrong
- AdapTable does not provide an agent of its own
- The Skills work with coding agents you already use, e.g. Cursor, Claude Code, VS Code with Copilot, OpenAI etc.
Once these Skills are installed, your agent sets up, changes, upgrades and debugs AdapTable for you, in far less time than doing that work by hand.
The Skills follow the AdapTable version installed in the project, so the agent works with the release you are on
## What we provide
AdapTable provides 2 AI resources: **Skills** (what your agent follows) and an **MCP Server** (what those Skills call):
| Resource | What it is |
| ----------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- |
| [Agent Skills](https://www.adaptabletools.com/docs/getting-started-ai-skills/index.md) | Setup, development, upgrade and debug; install and then give your agent a command |
| [MCP server](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md#mcp-mode) | The tool the Skills call to ensure all works quickly and efficiently |
- If the MCP Server is not installed, the Skill runs the [CLI](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md) — the same package — and still does the work
- You never need to call the MCP Server yourself
## How it works
## Get started
### Install the AdapTable AI resources
Installing AI in AdapTable is straightforward.
There are 3 steps in the project folder, then 1 question to confirm it worked.
Then do either step 4 or step 5.
{`npx skills add https://www.adaptabletools.com`}
The four Skills are installed into the folder your agent reads.
Commit them so the whole team has them.
#### Cursor
- Skills: `.agents/skills/` in the project
- Session-start file: `AGENTS.md`, or a `.cursor/rules/*.mdc` rule
- Without prompts: `npx skills add https://www.adaptabletools.com -a cursor -y`
#### Claude Code
- Skills: `.claude/skills/` in the project
- Session-start file: `CLAUDE.md`
- Without prompts: `npx skills add https://www.adaptabletools.com -a claude-code -y`
#### VS Code (Copilot)
- Skills: `.agents/skills/` in the project
- Session-start file: `AGENTS.md`
- Without prompts: `npx skills add https://www.adaptabletools.com -a github-copilot -y`
#### Codex
- Skills: `.agents/skills/` in the project
- Session-start file: `AGENTS.md`
- Without prompts: `npx skills add https://www.adaptabletools.com -a codex -y`
Paste this block into the project once, and commit it.
The agent reads that file at the start of each session, so you do not paste it again.
Use `AGENTS.md` (Codex, Cursor, VS Code etc.), `CLAUDE.md` (Claude Code) or `.cursor/rules/*.mdc` file (Cursor):
This is what makes the agent reach for the Skills reliably, before it has opened any of them.
Ask your agent:
Which AdapTable Skills do you have, and what AdapTable version is this project on?
It should name the four Skills and the AdapTable version from your `package.json`.
Install the MCP server so the Skills can call it.
Add it to the agent's configuration — `.cursor/mcp.json` in Cursor, or `.mcp.json` in Claude Code:
See [MCP mode](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md#mcp-mode) for VS Code, Codex and the other clients.
If you have skippled the previous step, the Skill runs the CLI instead, and the work still gets done.
It runs the CLI through `npx`.
If `npx` is blocked or slow, pin the package in the project so the run starts instantly and the version stays fixed:
{`npm i -D @adaptabletools/adaptable-mcp`}
See [Install and run](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md#install-and-run) for the details.
#### Requirements
There are 2 main requirements:
- Node 22.20 or later for the `skills` installer (the CLI itself needs Node 20 or later)
- An agent that reads Agent Skills, e.g. Cursor, Claude Code, VS Code (GitHub Copilot), OpenAI Codex etc.
## Stays on your machine
Nothing here is a cloud service.
All the pieces are files in your project and a tool that runs on your computer:
| Piece | Runs where | Talks to |
| --------------------------------------------------- | ----------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| Agent Skills | Files in your project, read by your agent | Nothing: they are written instructions |
| MCP if installed / `adaptable` CLI | Your machine | adaptabletools.com, to download the docs for your version; npm, to download itself and, only if needed, an AdapTable package |
| DevTools In-page API | The browser page your Grid runs in | The agent's browser tool on your machine |
- No new agent to buy, host or approve — developers keep the tools you already standardised on
- Nothing is sent to AdapTable Tools: no telemetry, no account, no API key
- Nothing is added to the application: the Skills and the CLI are development tools only
- Licence keys go into environment variables, never into source code, and the Skills never repeat a key back
- Your agent's own data policy is unchanged; these tools add no destination for your code
- Support-case and feedback files are written locally, scrubbed by default, and only a person sends them
Every network call is a download:
- adaptabletools.com: docs for your AdapTable version, kept in cache after 1st time (plus Skills' version no)
- npm: the CLI itself (and, if project has no AdapTable package to check, a copy of one)
- Nothing is uploaded when using the AdapTable AI tools
- Your source code, Initial State, row data, licence keys and prompts never leave the machine
See [Offline and restricted networks](https://www.adaptabletools.com/docs/technical-reference-ai-cli/index.md#offline-and-restricted-networks) for the cache folder, proxies and working with no network
## AG Grid Skills
The AdapTable Skills, MCP Server and CLI cover AdapTable.
Column definitions, AG Grid modules, row models and the Grid API belong to AG Grid.
- Install AG Grid's [Skills](https://www.ag-grid.com/javascript-data-grid/skills/) (`ag-dev`, `ag-update`) and, if you use MCP, its [MCP Server](https://www.ag-grid.com/javascript-data-grid/mcp-server/) alongside
- The AdapTable Skills hand AG Grid work over to them
## Agents without a shell
Every page of these docs is also published as markdown:
- `https://www.adaptabletools.com/llms.txt` lists every page
- `llms-full.txt` holds them all in one file
- `/index.md` after any page URL returns that page
The Skills tell an agent to read those pages rather than recall an older API.
- If a Skill misled the agent, a check missed something or missing docs page, ask agent for `ADAPTABLE_FEEDBACK.md`
- Review it and send it through a [support ticket](https://www.adaptabletools.com/support/adaptable-support-ticket) — that is how we decide what the Skills and the CLI do next.
### FAQ
**Does any of this send my code anywhere?**
No. Every network call is a download — docs from adaptabletools.com, packages from npm — and nothing is uploaded. See [Stays on your machine](#stays-on-your-machine).
**Does it work offline?**
Yes, once the docs for your version have been downloaded: the CLI works from its cache, and can be prepared ahead of time for a machine with no network. See [Offline and restricted networks](https://www.adaptabletools.com/docs/technical-reference-ai-cli/index.md#offline-and-restricted-networks).
**What does it cost?**
Nothing. There is no account, no API key and no extra licence.
**Which agents does it work with?**
Any agent that reads Agent Skills: Cursor, Claude Code, VS Code (GitHub Copilot), OpenAI Codex and most others.
**Do I need the MCP server?**
Install the Skills or nothing works, and install the MCP server as well. When it is configured, the Skill calls its tools, which is the easiest path. When it is not, the Skill runs the CLI and still finishes the job. You do not call MCP yourself.
**Do I need all four Skills?**
The install adds all four, and each is used only for its own job. `adaptable-dev` is the one you will use most.
**Does it work with AdapTable 21, 22 or 23?**
Yes. The docs are read for your version, and `adaptable-upgrade` plans upgrades from 21 or later. The checks run with an AdapTable 24 or later engine, which the CLI downloads when the project does not have one.
**Can I run the checks without an agent?**
Yes — in a terminal, a pre-commit hook or CI. See [adaptable CLI](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md).
**Does it work in a monorepo?**
Yes. The tools find every AdapTable project and work in the one the task is about. See [Install and run](https://www.adaptabletools.com/docs/getting-started-ai-cli/index.md#install-and-run).
**Does the MCP server have prompts or resources?**
No — tools only. The workflows live in the Skills, so they work in every agent, with or without MCP.