VectorClaw
VectorClaw is a local MCP server for setting up, controlling, and querying an Anki Vector robot through AI assistants.
Run guided setup: validate Python/SDK, write config, check connectivity, and run smoke tests.
Make Vector speak and play animations.
Drive straight/turn, drive off/on charger, and emergency stop.
Set head angle and lift height.
Capture camera images, scan, and use perception tools for faces/objects/vision reset.
Display custom images on Vectorβs face screen.
Interact with Vectorβs cube: dock, pickup, drop, roll.
Query robot pose, battery/charging status, charger status, touch sensor, and proximity sensor.
Provides tools for controlling an Anki Vector robot over a local Wire-Pod connection: making it speak and play animations, driving and docking, setting head angle and lift height, capturing camera images and scanning for faces/objects, interacting with the cube accessory, and reading pose, battery, charger, touch, and proximity status.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@VectorClaw tell Vector to say 'hello' and wave"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
π€ VectorClaw
Give your AI assistant a body.
VectorClaw is an MCP server that exposes Anki Vector robot capabilities as tools for AI assistants like OpenClaw β bridging natural language to real-world robot actions over a fully local, cloud-free stack.
Architecture
βββββββββββββββ stdio MCP ββββββββββββββββββββ gRPC/WiFi βββββββββββ
β AI Agent β βββββββββββββββ β vectorclaw-mcp β βββββββββββββββ β Vector β
β (OpenClaw) β β (Python 3.11+) β β Robot β
βββββββββββββββ ββββββββββββββββββββ βββββββββββAll communication is local-only β no cloud dependency at runtime (Wire-Pod setup path).
See Security Architecture for the full trust model.
Related MCP server: IoT Device Management MCP Server
Table of Contents
Current Status
v1.0.0 Released Β· 2026-03-01
16 tools verified β on hardware | 7 tools experimental β οΈ
Known limitations:
vector_drive_on_chargerβ requires the charger to be in Vector's recently-observed world model; callvector_scanfirst to give Vector a chance to locate the charger before attempting to dock (observed: invoking while charger is unobserved causes the command to time out without any robot motion)Perception tools (
vector_find_faces,vector_list_visible_faces,vector_face_detection,vector_list_visible_objects) β return empty results when vision modes are not active; vision modes are enabled automatically at connect but will be disabled for the remainder of the session after avector_vision_resetcall (reconnect to restore)Idle behaviors β Vector's autonomous animations can overlap with commanded behaviors; use one-command-at-a-time sequencing and wait for completion before issuing the next command
See ROADMAP.md for the full milestone plan.
Quickstart
Requirements: Python 3.11+ Β· Wire-Pod running Β· Vector on local WiFi
For the complete walkthrough (Wire-Pod install, robot auth, WiFi config, troubleshooting) see docs/SETUP.md.
Guided Setup (recommended for new users)
The vectorclaw-setup wizard handles configuration, SDK validation, connectivity
check, and a smoke test in one go:
pip install vectorclaw-mcp
vectorclaw-setupYou will be prompted for your robot's serial number and optional IP address. On success you'll see a clear SETUP PASSED message and the next-steps command. On failure every step includes an exact remediation hint.
See docs/OPENCLAW_SETUP_SKILL.md for full details.
Manual Setup
Step 1 β Install VectorClaw
pip install vectorclaw-mcpStep 2 β Configure Vector SDK
Wire-Pod is the canonical self-hosted server for Vector. Install the SDK distribution and run the one-time auth wizard:
pip install wirepod_vector_sdk
python -m anki_vector.configureNote:
wirepod_vector_sdkinstalls under theanki_vectorPython namespace, so all imports and CLI commands useanki_vector.
The standalone anki_vector package requires working DDL cloud servers and is brittle on modern Python runtimes.
pip install "vectorclaw-mcp[legacy]"
python -m anki_vector.configurePrefer wirepod_vector_sdk for reliable, cloud-independent operation.
Step 3 β Set environment variables
export VECTOR_SERIAL="your-robot-serial" # required β printed on underside of robot
export VECTOR_HOST="192.168.x.x" # optional β auto-discovered if omitted
export VECTOR_REPEAT_GUARD_SECONDS="10" # optional β refuse identical actuating calls within N s (0 = off)Step 4 β Run the server
vectorclaw-mcp
# or
python -m vectorclaw_mcpMCP Client Configuration
Add the following block to your mcporter.json (or equivalent MCP client config).
With uvx (recommended β no prior install needed)
{
"mcpServers": {
"vectorclaw": {
"command": "uvx",
"args": ["vectorclaw-mcp"],
"env": {
"VECTOR_SERIAL": "your-serial-here"
}
}
}
}With pip install (if installed locally)
{
"mcpServers": {
"vectorclaw": {
"command": "vectorclaw-mcp",
"env": {
"VECTOR_SERIAL": "your-serial-here"
}
}
}
}Available Tools
Tool | Category | Description | Status |
| ποΈ Speech | Make the robot speak text aloud | β |
| π Expression | Play a named animation | β οΈ |
| ποΈ Motion | Drive the robot off its charger | β |
| ποΈ Motion | Drive straight and/or turn in place | β |
| ποΈ Motion | Drive Vector back onto its charger | β οΈ |
| ποΈ Motion | Stop all motion immediately | β |
| π¦Ύ Actuation | Set head angle (β22Β° β 45Β°) | β |
| π¦Ύ Actuation | Set lift height (0.0 β 1.0) | β |
| π Perception | Capture image from front camera | β |
| π Perception | One-shot image capture | β |
| πΌοΈ Display | Display custom image on face screen | β |
| π Perception | Head scan for environment | β |
| π Perception | Scan for faces | β οΈ |
| π Perception | List currently visible faces | β οΈ |
| π Perception | Get face detection summary | β οΈ |
| π Perception | List currently visible objects | β οΈ |
| π² Interaction | Interact with cube (dock/pickup/drop/roll) | β οΈ |
| π Perception | Disable all vision modes | β |
| π Sensing | Get current position and orientation | β |
| π Status | Get battery level and charging status | β |
| π Status | Get charger connection state | β |
| π Status | Get touch sensor state | β |
| π Status | Get proximity sensor reading | β |
Status legend: β Verified on hardware | β οΈ Experimental (requires specific preconditions; see API Reference)
β οΈ Charger prerequisite:
vector_driverequires the robot to be off the charger. Callvector_drive_off_chargerfirst, or setVECTOR_AUTO_DRIVE_OFF_CHARGER=1for automatic undocking.
See docs/MCP_API_REFERENCE.md for full parameter details and response schemas.
Contributing
πΏ Branch: branch off
dev, use<type>/<short-description>naming (e.g.fix/vector-face-payload,feat/vector-scan)π§ͺ Tests: add or update tests under
tests/; all tests use the mocked SDK β no hardware requiredβ CI: Python 3.11 is required and must pass; Python 3.12 is experimental/informational β run
pytest tests/ -vlocally before opening a PRπ€ Hardware: if your change touches a tool or connection layer, record a smoke-test run in Hardware Smoke Log following the Hardware Test Playbook
π― PR scope: keep PRs focused β separate docs, feature, and refactor changes to reduce merge-conflict risk with parallel lanes
Docs Map
π οΈ Setup & Runtime
Document | Description |
Wire-Pod install, robot auth, WiFi, SDK config, troubleshooting | |
Common runtime failures, smoke baseline, and escalation path | |
Supported Python versions and CI policy |
π‘ API & SDK
Document | Description |
MCP tool signatures, parameters, response schemas | |
Full SDK capability catalog | |
Now/Later/Skip decision table for future tools |
π¬ Hardware Validation
Document | Description |
Repeatable on-robot validation protocol and PR checklist | |
Running record of real-world smoke tests | |
Which tools require undocked state |
π Security
Document | Description |
Threat model, credential handling, input validation, network posture |
License
MIT β see LICENSE.
Available Tools
24 toolsvector_animateB
Play a named animation on the robot.
Inputs: animation_name (required).
vector_capture_imageA
Capture a single camera frame via camera.capture_single_image.
Inputs: no named parameters.
vector_charger_statusB
Return charger and battery state.
Inputs: no named parameters.
vector_cubeC
Interact with Vector's cube accessory.
Inputs: action (required).
vector_driveB
Move the robot straight and/or turn it in place.
Inputs: speed, angle_deg, distance_mm.
vector_drive_off_chargerA
Drive the robot off its charger.
Inputs: no named parameters.
vector_drive_on_chargerA
Drive Vector back onto its charger with a configurable timeout. Returns immediately (already_on_charger: true) if Vector is already docked. Requires the charger to be in Vector's recent world model; use vector_scan first if the charger may not have been recently observed. Attempts a motor stop as a best-effort fallback if the maneuver times out.
Inputs: timeout_sec.
vector_emergency_stopA
Immediately stop all Vector motors.
Inputs: no named parameters.
vector_faceB
Display a custom image on the robot's face screen (144Γ108).
Inputs: image_base64 (required), duration_sec.
vector_face_detectionA
Return a summary of currently visible faces (no raw image data).
Inputs: no named parameters.
vector_find_facesB
Make Vector actively search for faces in the environment.
Inputs: no named parameters.
vector_headA
Set Vector's head angle in degrees. Input is clamped to the safe range (-22.0 to 45.0 degrees).
Inputs: angle_deg (required).
vector_liftA
Set Vector's lift/arm height (0.0 = lowest, 1.0 = highest). Input is clamped to the valid range 0.0β1.0.
Inputs: height (required).
vector_list_visible_facesB
Return the list of faces currently visible to Vector.
Inputs: no named parameters.
vector_list_visible_objectsB
Return the list of objects currently visible to Vector.
Inputs: no named parameters.
vector_lookB
Capture an image from the robot's front camera.
Inputs: no named parameters.
vector_poseA
Get the robot's current position and orientation.
Inputs: no named parameters.
vector_proximity_statusA
Return proximity sensor reading from Vector's front IR sensor.
Inputs: no named parameters.
vector_sayB
Make the Anki Vector robot speak text aloud.
Inputs: text (required).
vector_scanA
Make Vector look around in place to scan the environment.
Inputs: no named parameters.
vector_setupA
Run the VectorClaw setup wizard: validate Python compatibility, install the Vector SDK if missing, write OpenClaw config, check robot connectivity, and run smoke tests. Returns a structured PASS/FAIL report with actionable remediation steps. Call this tool first when setting up VectorClaw for a new user.
Inputs: host, serial, run_smoke, install_sdk, +2 more.
vector_statusA
Get robot status (battery level, charging state, etc.).
Inputs: no named parameters.
vector_touch_statusA
Return touch-sensor reading from Vector's back capacitive sensor.
Inputs: no named parameters.
vector_vision_resetB
Disable all active vision modes via vision.disable_all_vision_modes.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogβ¦
TDQS
Scored across 24 tools
Several tools have effectively the same purpose: vector_look and vector_capture_image both capture a camera frame, vector_status and vector_charger_status both report battery/charging state, and vector_face_detection overlaps heavily with vector_list_visible_faces. An agent could easily select the wrong tool despite the descriptions.
Most tools follow a vector_ + verb_noun pattern (drive, find_faces, capture_image), and the shared prefix makes the set feel unified. A few noun-style names like vector_pose, vector_status, vector_cube, and vector_face deviate slightly, but they are still predictable and readable.
24 tools is at the heavy end for a single server, and redundant pairs inflate the count. However, the robot-control domain legitimately needs many operations for driving, sensors, vision, and display, so the scope is borderline rather than excessive.
The set covers core Vector workflows: movement, docking, camera, speech, animation, face/object detection, head/lift control, sensor status, and setup. Lacking explicit audio capture or a dedicated stop-speaking command are minor gaps that don't create dead ends.
Maintenance
Related MCP Connectors
- DNS-DATA RemoteDeprecatedcom.dns-data
Connect AI assistants to authorized Windows and macOS devices for files, commands, and management.
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
Memory, reflection and safety for AI agents. Hosted or local, no vector database.
Turns a phone into a camera+Bluetooth remote so AI assistants can see and control any PC.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables natural language control of Reachy Mini robots for tasks like head movement, camera capture, and performing choreographed dances or emotions. It includes local, low-latency text-to-speech capabilities with synchronized head animations for lifelike interaction.21-
- AlicenseAqualityDmaintenanceEnables registration, monitoring, and control of IoT devices via AI agents, with local storage and no cloud API key required.9MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to control Home Assistant via natural language, including device control, automation management, and system monitoring.MIT
- AlicenseNot gradedqualityCmaintenanceConnects embodied devices (like StackChan, Raspberry Pi, ESP32) to AI via MCP protocol, enabling motion control with zero API cost, no PC required, and fully self-hosted.MIT