Overview API + CLI + MCP

Agent & Programmatic Access

CenterOS exposes a unified Action Layer with 39 actions across 9 domains. Access it three ways: REST API, CLI, or MCP for AI tools.

Updated Apr 2026 39 actions 9 categories
Looking for the complete developer toolkit? See CenterOS -- AI-Powered Dev Tools (CLI + MCP + API) for the full overview with quick start, MCP capabilities, example interactions, and supported hardware. →

Why Programmatic Access Matters

Robotics workflows are inherently multi-step: collect data, label it, train a model, evaluate, deploy to a fleet, monitor. Each step involves different tools, teams, and timelines. Programmatic access lets you automate these workflows end-to-end:

  • CI/CD for robots -- Trigger training jobs, run evaluations, and deploy models from your pipeline.
  • AI-assisted operations -- Let Claude or Cursor manage your platform using natural language.
  • Custom dashboards -- Build internal tools that pull live data from the platform API.
  • Fleet orchestration -- Script deployments, monitor health, and respond to alerts automatically.
  • Data pipelines -- Upload datasets, run annotation tasks, and track quality metrics programmatically.

Three Ways to Access

REST API

Direct HTTP access to all 31 platform actions. Use from any language, any platform. JWT and PAT authentication with fine-grained scopes.

curl / httpx / fetch →

CenterOS CLI

Terminal-first workflow tool with rich tables and formatted output. Also usable as a Python SDK in scripts and notebooks.

pip install centeros-cli →

MCP Server

Connect the platform to AI assistants -- Claude Code, Claude Desktop, Cursor, Windsurf. All actions become AI tools via natural language.

pip install centeros-mcp →

Getting Started

Regardless of which access method you choose, the setup flow is the same:

1

Create a platform account

Sign up at platform.roboticscenter.ai or ask your org admin to invite you.

2

Get a Personal Access Token

Install the CLI with pip install centeros-cli, then run centeros login followed by centeros pat create --name "my-token" --scopes "*". Store the frl_... token securely.

3

Choose your access method

Use the REST API for direct HTTP integration, the CLI for terminal workflows or Python scripts, or the MCP server for AI-assisted operations.

4

Start building

All three methods access the same 39 actions with the same authentication and permissions. Mix and match as needed.

shell
# Step 1: Install the CLI
$ pip install centeros-cli

# Step 2: Login and create a PAT
$ centeros login --username your-username --password '***'
$ centeros pat create --name "my-agent" --scopes "*"
# PAT created: frl_abc123...   (save this!)

# Step 3: Use it anywhere
$ export CENTEROS_TOKEN=frl_abc123...
$ centeros datasets list               # CLI
$ curl .../api/actions -H "Authorization: Bearer $CENTEROS_TOKEN"  # API

The Action Layer

All three access methods share the same backend: the Action Layer. It is a unified registry of 39 executable actions organized into 9 categories:

  • Datasets (6) -- list, get, delete, upload, signed URL, confirm upload
  • Training (6) -- list jobs, create, status, launch, catalog, estimate cost
  • Models (4) -- list, register, promote lifecycle, get lineage
  • Fleet (6) -- summary, list robots, robot state, register, deploy model, emergency stop
  • Simulation (4) -- list scenes, create scene, launch sim, sim status
  • Processing (4) -- list pipelines, create pipeline, run status, cancel run
  • Annotations (4) -- list tasks, create task, metrics, submit for review
  • Organizations (3) -- list orgs, members, invite
  • Stats (2) -- platform statistics, system health

Every action has a name, description, JSON input schema, and required scopes. Discovery is automatic: GET /api/actions returns the full catalogue. See the Platform API Reference for complete details.

How to Choose

Use CaseBest ChoiceWhy
CI/CD pipeline automationCLI or REST APISet CENTEROS_TOKEN in CI, run centeros commands or curl.
Python notebook / scriptPython SDKfrom centeros import CenterOSClient gives you typed methods.
Non-Python integrationREST APIStandard HTTP/JSON works with any language.
AI-assisted developmentMCP ServerClaude, Cursor, or Windsurf call platform actions via natural language.
Interactive terminal workCLIRich tables, formatted output, command completion.
Custom dashboard / frontendREST APIFetch data from your frontend with Bearer auth.

Dive in

API Reference Full REST API with all 39 actions, schemas, and curl examples → CLI Guide Install centeros-cli and manage everything from the terminal → MCP Integration Connect Claude, Cursor, or Windsurf to the platform → Hardware SDK Connect physical robots with pip install roboticscenter →