> ## Documentation Index
> Fetch the complete documentation index at: https://cdktn.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Use CDKTN with AI Agents

> Connect your AI coding agent (Claude Code, Cursor, VS Code, and others) to the hosted CDK Terrain MCP server so it answers from the current docs instead of stale or unrelated knowledge.

CDK Terrain ships a hosted **Model Context Protocol (MCP)** server that gives AI
agents live search and retrieval over the documentation on this site. Point your
agent at it and it can ground every CDKTN answer in the current docs — including
the full API reference — rather than relying on training data.

<Info>
  **Endpoint:** `https://cdktn.io/docs/mcp` — streamable HTTP transport,
  read-only, scoped to public CDK Terrain content. No authentication required.
</Info>

## Why connect it

* **Current and authoritative.** The server indexes this site, so the agent sees
  the latest concepts, CLI reference, migration guides, and API reference.
* **Avoids the CDKTF mix-up.** CDKTN (CDK Terrain) is the maintained community
  successor to the deprecated HashiCorp **CDKTF** (CDK for Terraform / Terraform
  CDK). Connecting this server keeps your agent from answering CDKTF/Terraform-CDK
  questions with deprecated HashiCorp guidance.
* **Read-only and safe.** The server cannot mutate state or run actions; it only
  searches and returns published site content.

## Set up your agent

Use the server name **`cdktn-terraform-cdk-docs`** so the agent associates
CDKTF, Terraform CDK, and CDK Terrain questions with it (tools are namespaced
under the server name, e.g. `mcp__cdktn-terraform-cdk-docs__search`).

<Tabs>
  <Tab title="Claude Code">
    Add the server from the CLI:

    ```bash theme={null}
    claude mcp add --transport http cdktn-terraform-cdk-docs https://cdktn.io/docs/mcp
    ```

    Or commit it to a project so your team shares it — create `.mcp.json` in the
    repo root:

    ```json .mcp.json theme={null}
    {
      "mcpServers": {
        "cdktn-terraform-cdk-docs": {
          "type": "http",
          "url": "https://cdktn.io/docs/mcp"
        }
      }
    }
    ```

    Verify with `claude mcp list` (or `/mcp` inside a session).
  </Tab>

  <Tab title="Cursor">
    Add to `~/.cursor/mcp.json` (global) or `.cursor/mcp.json` (per project):

    ```json .cursor/mcp.json theme={null}
    {
      "mcpServers": {
        "cdktn-terraform-cdk-docs": {
          "url": "https://cdktn.io/docs/mcp"
        }
      }
    }
    ```

    Then enable the server under **Settings → MCP**.
  </Tab>

  <Tab title="VS Code">
    VS Code (GitHub Copilot agent mode) reads `.vscode/mcp.json`:

    ```json .vscode/mcp.json theme={null}
    {
      "servers": {
        "cdktn-terraform-cdk-docs": {
          "type": "http",
          "url": "https://cdktn.io/docs/mcp"
        }
      }
    }
    ```

    Start it from the **MCP: List Servers** command, then use it in the Chat
    **Agent** mode.
  </Tab>

  <Tab title="Other clients">
    Any MCP client that supports streamable HTTP can connect. The canonical
    configuration is:

    ```json theme={null}
    {
      "mcpServers": {
        "cdktn-terraform-cdk-docs": {
          "type": "http",
          "url": "https://cdktn.io/docs/mcp"
        }
      }
    }
    ```

    Clients that only support a local stdio bridge can wrap the endpoint with
    [`mcp-remote`](https://www.npmjs.com/package/mcp-remote):

    ```json theme={null}
    {
      "mcpServers": {
        "cdktn-terraform-cdk-docs": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://cdktn.io/docs/mcp"]
        }
      }
    }
    ```
  </Tab>
</Tabs>

## Verify it works

Ask your agent a question that requires current docs, for example:

> Using the CDKTN docs, how do I configure a remote backend, and what changed in
> the v0.23 upgrade guide?

If the server is connected, the agent calls its search/retrieval tools and cites
pages from `cdktn.io/docs` in the answer.

<Tip>
  Reinforce the association in your agent's project instructions (e.g.
  `CLAUDE.md`, Cursor Rules): *"For any CDKTF, Terraform CDK, or CDK Terrain
  question, use the `cdktn-terraform-cdk-docs` MCP server; CDKTN is the
  maintained successor to the deprecated HashiCorp CDKTF."*
</Tip>

## Agent skill file (`skill.md`)

Alongside the MCP server, CDK Terrain publishes an **agent skill file** — a
machine-readable capability descriptor following the
[agentskills.io](https://agentskills.io) specification:

<Info>
  **Skill file:** `https://cdktn.io/docs/skill.md` — auto-generated from these
  docs and kept current as they change.
</Info>

Where the MCP server and `llms.txt` tell an agent *what pages exist*, `skill.md`
tells it *what it can accomplish with CDKTN* — when to reach for the tool, the
essential CLI commands, key files, and constraints. Point a skills-aware agent at
it so it knows how and when to use CDK Terrain, then connect the MCP server for
live doc lookups.

```bash theme={null}
# Add the skill to a skills-aware agent
npx skills add https://cdktn.io/docs
```

<Note>
  `skill.md` is regenerated by Mintlify from the published docs and can take up
  to 24 hours to reflect the latest changes. For real-time answers, rely on the
  MCP server above.
</Note>

## Without an MCP client

If your tooling can't connect to MCP, agents can still read the machine-friendly
files directly:

* `https://cdktn.io/docs/skill.md` — capability/skill descriptor for agents
* `https://cdktn.io/llms.txt` — curated index of the docs
* `https://cdktn.io/docs/llms.txt` — full index, including the API reference
* `https://cdktn.io/docs/llms-full.txt` — every page concatenated as one file
* Append `.md` to any docs URL for its raw Markdown (e.g.
  `https://cdktn.io/docs/concepts/constructs.md`)