Summary

A robotics API surface, not a robot driver.

Every tool call goes to the NWO cloud, so what you can actually do depends on what your account and hardware are wired to over the ROS2 bridge — the server itself is a typed front door. The useful shape here is breadth: mapping, planning, RL training and five sensor families under one connection, which is hard to assemble from separate integrations. Start with `nwo_edge_inference` if latency matters; the standard endpoint is the slower of the two.

What it is

A TypeScript MCP server that wraps the NWO Robotics API. It carries 77 tools covering mapping and localisation, motion planning, reinforcement-learning training, a set of non-camera sensors, and safety monitoring, so a model can plan and run robot work through tool calls instead of REST plumbing.

What you get

Persistent mapping and localisation — `nwo_initialize_slam` builds a named map, `nwo_localize` places the robot inside one from a single image
Open-vocabulary perception — `nwo_detect_objects_grounding` finds objects from a plain-language description, `nwo_identify_material` names what a surface is
Sensors a camera cannot give you — `nwo_query_thermal`, `nwo_query_mmwave`, `nwo_query_gas`, `nwo_query_acoustic`, `nwo_query_magnetic`, and `nwo_read_tactile` for the ORCA Hand's 576 taxels
Planning and execution — `nwo_plan_motion` for collision-aware paths, `nwo_task_planner` and `nwo_execute_subtask` for decomposing an instruction, `nwo_execute_behavior_tree` for hierarchical runs
Cloud reinforcement learning — `nwo_create_rl_env` creates a training environment, `nwo_train_policy` runs the training job
Two more inference paths: `nwo_sensor_fusion` folds sensor readings into an instruction, and `nwo_edge_inference` targets the low-latency edge endpoint

Requirements

Node.js 18 or newer and a build step: `npm install` then `npm run build`. Authentication is a single environment variable, `NWO_API_KEY`, from the NWO developer page; the free tier is documented at 100,000 calls per month. Endpoints are configurable through `NWO_API_BASE`, `NWO_EDGE_API`, `NWO_ROS2_BRIDGE`, `MQTT_BROKER` and `MQTT_PORT`, so the same server can point at the standard API, the edge deployment, or a ROS2 bridge for physical robots. A Dockerfile and a docker-compose stack with an MQTT broker ship with the package.

Setup effort

Build from source — clone the repository and build it