scry-connect 0.1.2

MCP server for ROS 2 — exposes robot capabilities as AI-callable tools

scry-connect

MCP server for ROS 2 — exposes your robot's topics, nodes, services, parameters, actions, diagnostics, and introspection tools to AI agents.

Pairs with the Scry Android app.

Install

Pick the path that matches how your robot is deployed. All three install the same scry-connect from PyPI — they just package it differently.

Option A — one-line installer (Docker or bare-metal, auto-detect)

curl -fsSL https://raw.githubusercontent.com/phaneron-robotics/scry-web/master/install.sh | bash

The script detects your ROS distro, picks Docker if available else pip, writes a systemd --user unit, starts the service, and prints a pairing QR. Re-running upgrades in place. Mode override: SCRY_INSTALL_MODE=docker bash or SCRY_INSTALL_MODE=pip bash.

Option B — sidecar Docker container

Drop into your existing docker-compose.yml:

services:
  scry-connect:
    image: ghcr.io/phaneron-robotics/scry-connect:${ROS_DISTRO:-jazzy}
    network_mode: host
    ipc: host
    pid: host
    restart: unless-stopped
    environment:
      - ROS_DOMAIN_ID=${ROS_DOMAIN_ID:-0}
    volumes:
      - scry_config:/home/scry/.config/scry
      - scry_audit:/var/log/scry

volumes:
  scry_config:
  scry_audit:

A turnkey reference compose with every tunable surfaced lives at docker/docker-compose.yml.

Tag	Resolves to
:humble, :jazzy, :kilted, :lyrical, :rolling	Per-ROS-distro images (multi-arch: amd64 + arm64)
:0.1.0-jazzy	Pinned version on jazzy
:latest	Most recent stable release on the LTS distro (jazzy today)

Option C — add scry-connect to your own robot image

If you already publish a robot Docker image, append one line to your Dockerfile and run scry-connect alongside your existing nodes:

RUN pip install scry-connect

No second ROS install required — scry-connect uses the rclpy that's already in your image.

Option D — bare-metal pip (no Docker)

source /opt/ros/$ROS_DISTRO/setup.bash
pip install --user scry-connect
scry-connect

A reference systemd user unit is written automatically by Option A in pip mode.

Prereleases (alpha / beta / rc) live on TestPyPI:

pip install -i https://test.pypi.org/simple/ scry-connect

Run

# Defaults to 0.0.0.0:5339 in open mode (LAN-only, no token).
scry-connect

# Common overrides
scry-connect --port 5339 --log-level INFO
scry-connect --token              # token mode; prints QR for the phone
scry-connect --mtls               # client-cert mode
scry-connect --skip-ros-check     # smoke test without rclpy

Endpoints

Path	Method	Description
/mcp	POST/GET	MCP JSON-RPC (Streamable HTTP)
/stream?topic=/odom&rate=10	GET	SSE topic stream
/health	GET	Health check

Tools exposed

22 tools grouped into: Topics, Nodes, Services, Parameters, Actions, Diagnostics, Logs, Introspection, System. See docs/mcp-tools-reference.md.

Config (env vars)

Variable	Default	Description
SCRY_HOST	0.0.0.0	Bind address
SCRY_PORT	5339	HTTP port
SCRY_RATE_LIMIT	10	Tool calls per second
SCRY_LOG_LEVEL	INFO	Log level
SCRY_AUTH_MODE	open	open | token | mtls
SCRY_TOKEN	(unset)	Force token mode with this shared token
SCRY_PUBLIC_INTERNET	0	In open mode, accept non-LAN callers
SCRY_REQUIRE_DEADMAN	0	Writes need /scry/enable heartbeat
SCRY_AUDIT_LOG	(unset)	Append JSONL audit trail to this path
SCRY_MDNS	1	Advertise _scry._tcp via Zeroconf
ROS_DOMAIN_ID	(inherited)	ROS 2 domain ID
RMW_IMPLEMENTATION	(inherited)	DDS/RMW implementation

Development

Setup

# Editable install + dev dependencies
pip install -e ".[dev]"

# Install pre-commit hooks (run once per fresh clone)
pip install pre-commit
pre-commit install

After pre-commit install, every git commit automatically runs ruff (auto-fix) + ruff format on the staged Python files. If anything was fixed, the commit is rejected — git add the fixes and commit again.

Manual lint commands

# Run pre-commit against every file (not just staged)
pre-commit run --all-files

# Run ruff directly
ruff check scry_connect tests          # report issues
ruff check --fix scry_connect tests    # auto-fix what can be fixed
ruff format scry_connect tests         # apply formatter

Tests

pytest tests/

Note: many tests require a working ROS 2 environment (rclpy). In CI we only run lint + build; tests run on a real robot or local ROS 2 install.

CI / release pipeline

Workflow	Trigger	What it does
CI	PR to master, push to master, manual	ruff check + format check, python -m build (sdist + wheel)
Release	Tag matching v*, manual	Build, publish to PyPI / TestPyPI (Trusted Publishing — no API key needed), push to ghcr.io/phaneron-robotics/scry-connect, create GitHub Release
Dependency Graph	Auto-managed	Tracks third-party deps for security alerts

To cut a release: bump version in pyproject.toml, merge the bump PR, then tag the merge commit. Prerelease tags (v0.1.0-beta.1) route to TestPyPI + a prerelease GitHub Release. Stable tags (v0.1.0) route to PyPI proper and the Docker image gets the latest tag.

See docs/PYPI_SETUP.md for the one-time PyPI Trusted Publishing setup.