Krater MCP Server

Krater tools inside any MCP client Krater MCP is a public Model Context Protocol (MCP) server that exposes Krater's AI tools — chat models, image and video generation, audio, text utilities, and we...

Krater tools inside any MCP client

Krater MCP is a public Model Context Protocol (MCP) server that exposes Krater’s AI tools — chat models, image and video generation, audio, text utilities, and web search — to external clients like Claude, ChatGPT, Cursor, and any other MCP-compatible app. Tool calls spend credits from your Krater plan, so one subscription covers both the app and your external clients.

Endpoint: https://api.krater.ai/v1/mcp

Requirements

• Ultra or Max plan. Krater MCP uses the same entitlement as the REST API.
• An MCP client that supports remote servers with OAuth sign-in. Current versions of Claude, ChatGPT, Cursor, and most MCP-compatible tools do.

If your client can’t run an OAuth flow, a kr_live_ API key works instead — create one in Settings → API and pass it as the Bearer token.

Connect

1. Add a new MCP server in your client with the URL https://api.krater.ai/v1/mcp.
2. The client opens a Krater sign-in page. Log in with your Krater account.
3. Review the consent screen — it shows what the app is allowed to do (use Krater tools for you and spend your plan credits) — and click Allow.
4. The client fetches the tool list and you’re done. Krater’s 20 tools are now callable inside that app.

A typical client config looks like:

{
  "mcpServers": {
    "krater": {
      "url": "https://api.krater.ai/v1/mcp"
    }
  }
}

With an API key instead of OAuth, send Authorization: Bearer kr_live_... on every request.

The server speaks standard MCP over HTTP (JSON-RPC 2.0, POST). Protocol versions 2025-03-26, 2025-06-18, and 2025-11-25 are accepted; everything else negotiates down to 2025-06-18.

The 20 tools

Async tools

generate_video and generate_music return a job_id immediately instead of blocking. Poll get_video_status or get_music_status with that ID until the job completes — the server tells clients this in its handshake instructions.

Input notes

• transcribe_audio and change_voice take public HTTPS URLs, not file uploads.
• summarize_text needs either text or url.
• humanize_text accepts a tone option.
• chat_completion uses provider/model-style IDs — call list_models first.

Credits and limits

• Billing: every tool call runs the same code path as the matching REST API endpoint, so it bills identically — the picked model’s real usage at the standard credit multiplier. Image generation on Nano Banana is about 10 credits, TTS 10 credits per 1,000 characters, video bills per second per model — same numbers you see in the app.
• Rate limit: 60 requests per minute per connection.
• Daily cap: 5,000 credits/day per connection by default, same as API keys.
• Hard caps: the API’s per-request ceilings apply — a single chat call can’t exceed 500 credits.
• Long text outputs are truncated at 100,000 characters.
• Audio results are saved to storage and returned as an audio_url you can download or pass to the next tool.

Manage and disconnect

Every connected app shows up in Settings → API as a Connected app entry named after the client. Revoke it there at any time — access stops immediately. kr_live_ API keys are created and managed on the same page.

Errors

Failures come back as tool errors carrying the API’s error message — INSUFFICIENT_CREDITS, PLAN_REQUIRED, RATE_LIMITED, and friends. The full list is in error codes.

Related

• Understanding credits — how Krater bills usage
• Plans & credits — Ultra and Max include API and MCP access
• API documentation — the same capabilities over raw REST
• Krater Keep — where generated media is stored

More in Features

• Additions: Apps, Personas & Prompts
• Personas: Specialist AI Assistants
• Scheduled Tasks
• Sharing & Projects
• API Access

See Krater pricing or contact support.