---
title: MCP | Adaption
description: Connect an MCP client to Adaption and use tools for datasets, synthetic data generation, and model training.
---

The Adaption Model Context Protocol (MCP) server lets AI assistants work with your Adaption datasets and AutoScientist runs. It is a hosted server, so there is nothing to deploy locally.

## Connect your client

The server is remote, so there is nothing to install or deploy. Use an Adaption plugin to configure the MCP connection and add skills that teach your assistant how to use the tools together.

Cursor

**Install**

1. Open **Customize** in the Cursor sidebar.
2. Search for **Adaption** and select **Install**.
3. Open **Configure** and set `ADAPTION_API_KEY` to your [Adaption API key](/introduction/create-api-keys/index.md).

**Verify**

Open **Cursor Settings → Tools & MCP** and confirm that `adaption` is connected. Then ask the agent to list your Adaption datasets.

**What the plugin adds**

Alongside the MCP connection, the plugin supplies three workflow skills: `adaption-dataset`, `adaption-invent`, and `adaption-training`. They guide the agent through multi-step dataset and AutoScientist workflows.

[Read the Cursor plugin guide](/tools/cursor-plugin/index.md).

Claude Desktop

**Install**

1. Download `adaption.mcpb` from the [Adaption Claude plugin releases](https://github.com/adaptionlabs/adaption-claude-plugin/releases).
2. Open the downloaded file to install it in Claude Desktop.
3. When prompted, enter your [Adaption API key](/introduction/create-api-keys/index.md).

The plugin supports Claude Desktop on macOS and Windows.

**Verify**

Restart Claude Desktop if prompted, then ask Claude to list your Adaption datasets.

**What the plugin adds**

Alongside the MCP connection, the plugin supplies three workflow skills: `adaption-dataset`, `adaption-invent`, and `adaption-training`. They guide Claude through multi-step dataset and AutoScientist workflows.

[Read the Claude plugin guide](/tools/claude-plugin/index.md).

Codex

An Adaption plugin for Codex is in development. Until it is available, connect Codex directly to the MCP server.

Set your API key in the environment:

Terminal window

```
export ADAPTION_API_KEY="pt_live_..."
```

Add the server to `~/.codex/config.toml`:

```
[mcp_servers.adaption]
url = "https://api.prod.adaptionlabs.ai/api/v1/mcp"
bearer_token_env_var = "ADAPTION_API_KEY"
```

Restart Codex and run `/mcp` to confirm that `adaption` is connected.

## Connect directly

Create an [Adaption API key](/introduction/create-api-keys/index.md), then configure your MCP client with:

- **URL:** `https://api.prod.adaptionlabs.ai/api/v1/mcp`
- **Authentication:** `Authorization: Bearer YOUR_ADAPTION_API_KEY`

For clients that use JSON configuration:

```
{
  "mcpServers": {
    "adaption": {
      "url": "https://api.prod.adaptionlabs.ai/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_ADAPTION_API_KEY"
      }
    }
  }
}
```

Replace `YOUR_ADAPTION_API_KEY` with your key. Do not commit a configuration file containing a real key.

## Dataset tools

- `list_datasets` — List datasets available to your organization.
- `get_dataset` — Get details about one dataset.
- `get_dataset_status` — Check processing status and progress.
- `get_dataset_evaluation` — Get dataset quality evaluation results.
- `get_dataset_export_links` — Get saved export URLs.
- `import_dataset` — Import from Hugging Face, Kaggle, or Google Sheets.
- `run_dataset_adaptation` — Estimate or launch dataset adaptation.
- `augment_dataset` — Add synthetic domain or general-purpose rows.
- `translate_dataset` — Translate sampled rows into target languages.
- `localize_dataset` — Localize sampled rows for country and language pairs.
- `combine_datasets` — Combine compatible datasets.

## Invent tools

- `list_invent_domains` — List valid domains and subdomains.
- `generate_dataset` — Estimate or generate a synthetic dataset from a description.

## AutoScientist tools

- `list_training_models` — List available base models.
- `list_autoscientist_runs` — List AutoScientist runs.
- `get_autoscientist_run` — Check one run’s status and results.
- `recommend_autoscientist_hyperparameters` — Get a recommended model and training configuration.
- `create_autoscientist_run` — Launch an AutoScientist training run.

## Operations that consume credits

Adaptation, augmentation, translation, localization, dataset generation, and AutoScientist training can consume credits. When a tool supports `estimate: true`, request an estimate and review it before launching the operation.

## Add workflow guidance

- [Install the Cursor plugin](/tools/cursor-plugin/index.md)
- [Install the Claude plugin](/tools/claude-plugin/index.md)