# Access Kyma Models from Claude Code with Kyma's MCP Server

Markdown version of https://kymaapi.com/blog/mcp-server-kyma, for agents and crawlers. Same content as the HTML page.

- Generated: 2026-10-09T21:45:05.594Z
- HTML page: https://kymaapi.com/blog/mcp-server-kyma
- Model catalog: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md

2026-04-12 · 2 min read

> **Update, 2026-09-14.** `kimi-k2.5`, suggested below for agentic tasks, was retired on 2026-09-11 and is no longer served on Kyma. Use `kimi-k2.6`.

> **Update, 2026-09-21.** `gemini-2.5-flash`, which this post suggested for long context, retires on 2026-10-20 and is no longer listed on Kyma. Use `gemini-3-flash`, which the `long-context` alias resolves to; the examples below now name it.

![MCP Server](/blog/mcp-server-hero.jpg)

## The Problem

Claude Code and Cursor are powerful. But sometimes you want to use a different model for a specific task — DeepSeek R1 for reasoning, Qwen 3.6 Plus for general quality, or Gemini 3 Flash for processing a massive codebase.

With Kyma's MCP server, you can access Kyma's active models directly from your AI coding tool. No switching windows, no separate API calls.

## Setup (2 Minutes)

### Claude Code

Add to `~/.claude/settings.json`:

```json
{
  "mcpServers": {
    "kyma": {
      "command": "npx",
      "args": ["@kyma-api/mcp-server"],
      "env": {
        "KYMA_API_KEY": "kyma-your-api-key"
      }
    }
  }
}
```

Restart Claude Code. Done.

### Cursor

Add to Cursor's MCP settings (Settings > MCP):

```json
{
  "kyma": {
    "command": "npx",
    "args": ["@kyma-api/mcp-server"],
    "env": {
      "KYMA_API_KEY": "kyma-your-api-key"
    }
  }
}
```

## What You Can Do

The MCP server exposes two tools:

### `chat` — Talk to Any Model

Ask your AI assistant to use Kyma for specific tasks:

- "Use kyma chat with deepseek-r1 to analyze this algorithm's time complexity"
- "Use kyma chat with qwen-3-32b to refactor this function"
- "Use kyma chat with gemini-3-flash to summarize this entire file"

### `list_models` — See Available Models

"Use kyma list_models to show me what's available"

Returns Kyma's active models with their IDs and providers.

## When to Use Which Model

| Task | Model | Why |
|------|-------|-----|
| Code review | `deepseek-v3` | Strong code understanding, best value |
| Reasoning/math | `deepseek-r1` | Chain-of-thought, 96% cheaper than o1 |
| General quality | `qwen-3.6-plus` | #1 most popular on Kyma |
| Fast iteration | `qwen-3-32b` | Ultra-fast responses |
| Large files | `gemini-3-flash` | 1M context window |
| Agentic tasks | `kimi-k2.5` | Best tool calling support |

## How It Works

The MCP server runs locally via `npx`. When your AI tool calls a Kyma tool, the server:

1. Receives the request via stdio
2. Forwards it to `https://kymaapi.com/v1/chat/completions`
3. Returns the response to your AI tool

Your API key stays local. No data is stored on Kyma's servers beyond what's needed for the request.

## Cost

With $0.50 free credit on signup — spendable on the [free tier](https://kymaapi.com/pricing#free-tier) — you get hundreds of MCP tool calls. A typical code review request costs about $0.002.

Get your API key at [kymaapi.com](https://kymaapi.com) and set it up in under 2 minutes.

## Links

- [MCP Server Guide](https://docs.kymaapi.com/guides/mcp-server) — detailed setup docs
- [Model Recommendations](https://docs.kymaapi.com/models/recommended) — pick the right model
- [Kyma API](https://kymaapi.com) — sign up and get started