Projects

Tool	Description
kite_project_list	List all projects in your workspace
kite_project_create	Create a new project. Params: name (string)
kite_project_get	Get full project details including file tree, config, and assets. Params: project_id

Robots

Tool	Description
kite_robot_list	List available robot models (built-in and custom). Optional: type (arm/humanoid/quadruped), source (builtin/custom)

Tasks

Tool	Description
kite_task_list	List task definitions (locomotion, balance, manipulation, hybrid). Reads from local registry.

Policies & Hardware

Tool	Description
kite_policy_list	List training policies (pi0, ACT, diffusion, etc.). Optional: robot_type
kite_hardware_list	List GPU hardware tiers with pricing
kite_hardware_estimate	Estimate training cost. Params: tier (gcp_gpu_t4/l4/a100/h100), steps (integer)

Datasets

Tool	Description
kite_dataset_import	Import a dataset from HuggingFace (repo ID) or S3 (s3:// URI). Params: input, project_id (optional)
kite_dataset_list	List available datasets. Optional: limit (default 20)
kite_dataset_validate_vla	Validate a dataset for VLA training compatibility. Params: dataset_id
kite_dataset_link	Link a dataset to a project. Params: project_id, dataset_id, role (train/eval)

Worlds

Tool	Description
kite_world_list	List generated worlds for a project. Params: project_id
kite_world_status	Check the status of a world generation job. Params: generation_id, project_id (optional)

Objects

Tool	Description
kite_object_list	List 3D objects in the library. Optional: project_id
kite_object_place	Place an object in the scene. Params: project_id, object_id, position [x,y,z], rotation [rx,ry,rz]

Training

Tool	Description
kite_train_preflight	Run pre-flight checks before starting training. Params: project_id, config (optional)
kite_train_start	Start a training job. Params: project_id, policy_id, hardware_tier, training_type (RL/VLA), config
kite_train_status	Get current status, metrics, and progress. Params: job_id
kite_train_logs	Get recent training logs. Params: job_id, tail (default 100)
kite_train_stop	Stop a running training job. Params: job_id
kite_train_artifacts	List downloadable artifacts (model, metrics, video). Params: job_id

Deployment

Tool	Description
kite_deploy_create	Create a deployment linking a trained checkpoint to a robot. Params: project_id, policy_id, checkpoint_ref, robot_type, mode (sim/real)
kite_deploy_list	List all deployments for the current user

Sessions

Tool	Description
kite_session_start	Start a simulator or robot session. Params: deployment_id, robot_id
kite_session_status	Get session state (loading/running/paused/terminated) and metrics. Params: session_id
kite_session_stop	Gracefully stop a running session. Params: session_id
kite_session_telemetry	Get per-step telemetry (joint positions, actions, rewards, latency). Params: session_id, last_n (default 50)

Agent orchestrator

Tool	Description
kite_agent_ask	Send a prompt to the Kite AI agent orchestrator. Routes to specialist agents (code, training, creator, motion, deployment, research, triage, action). Params: prompt, project_id (optional), conversation_id (optional)

Tip

The agent orchestrator is the most powerful tool. It classifies your intent, builds an execution plan across specialist agents, and returns structured responses with file proposals. Use it for complex multi-step workflows like "Build a pick-and-place task for the SO-ARM100 and start training."

Diagnostics

Tool	Description
kite_doctor	Check API connectivity, authentication status, and account info. No params required.

MCP Tools Reference