Documentation/

Developer

MCP Server

Connect Kite to Claude Code, Claude Desktop, ChatGPT, or any MCP-compatible AI assistant. One command, 30+ tools, full platform access.

Quick start

Quick start

Add Kite to Claude Code with a single command:

Terminal
claude mcp add --transport http kite https://mcp.kiteml.com/mcp

No install required. Claude connects to the hosted Kite MCP server over HTTP. All 30 tools are available immediately after restarting Claude Code.

Tip

You'll need an API key to authenticate. Generate one from Settings > API Keys in the Kite dashboard, then set it in your Claude Code MCP settings.

Overview

Overview

The Kite MCP server exposes the full Kite platform as native tool calls using the Model Context Protocol. It supports two transports:

Remote HTTP

Hosted at mcp.kiteml.com — no local install needed. One-liner setup.

Local stdio

Runs as a local subprocess via pip install. Works offline.

30+ tools

Projects, training, deployment, sessions, datasets, worlds, and more.

Agent orchestrator

Route prompts to 8 specialist agents via kite_agent_ask.

Authentication

Authentication

Generate an API key from Settings > API Keys in the Kite dashboard. The key starts with kite_ and is shown only once — store it securely.

For the hosted server, pass your key as a Bearer token. Claude handles this automatically when you configure it in your MCP settings:

~/.claude.json
{
  "mcpServers": {
    "kite": {
      "transport": "http",
      "url": "https://mcp.kiteml.com/mcp",
      "headers": {
        "Authorization": "Bearer kite_your_key_here"
      }
    }
  }
}

Claude Desktop

Claude Desktop

Claude Desktop also supports remote MCP servers. Add to your claude_desktop_config.json:

claude_desktop_config.json
{
  "mcpServers": {
    "kite": {
      "transport": "http",
      "url": "https://mcp.kiteml.com/mcp",
      "headers": {
        "Authorization": "Bearer kite_your_key_here"
      }
    }
  }
}

OpenAI / ChatGPT

OpenAI / ChatGPT

ChatGPT supports remote MCP servers via Apps & Connectors. Go to ChatGPT Settings > Apps & Connectors, add a new connector, and enter the server URL:

https://mcp.kiteml.com/mcp

Local installation

Local installation

Prefer to run locally? Install the CLI and register the stdio transport:

Terminal
pip install kite-cli[mcp]

claude mcp add kite -e KITE_API_KEY=kite_your_key_here -- kite-mcp

This runs the MCP server as a local subprocess. Works offline for local-only tools (task list, hardware estimates).

Verify it works

Verify it works

After setup, ask your AI assistant:

"Use kite_doctor to check connectivity"

List my projects

"List my Kite projects" — calls kite_project_list

Estimate costs

"Estimate 100k steps on an A100" — calls kite_hardware_estimate

Build a task

"Ask Kite to build a pick-and-place task" — calls kite_agent_ask

Run diagnostics

"Use kite_doctor to check connectivity" — calls kite_doctor

Environment variables

Environment variables

VariableDescription
KITE_API_KEYAPI key (recommended for MCP)
KITE_TOKENFirebase ID token (alternative)
KITE_API_URLAPI base URL (default: https://api.kiteml.com)

Get started with Kite ML.

Basic

Everything you need to start building with Kite.

Free
  • 1 free project
  • CPU processing
  • Simulation engines
  • Fast coding agent
  • Synthetic data generation
Popular

Hobbyist

Scale to production with cloud GPUs and the full agent.

$100/ user / month
  • Unlimited projects
  • GPU compute on demand
  • Simulation engines
  • Pro coding agent
  • Synthetic data generation

Startups

Tailored to your stack, your data, and your team.

  • Custom deployment pipelines
  • Bring your own datasets
  • Tailored workflows & integrations
  • Dedicated simulation environments
  • Forward deployed engineers