# Live Debugging MCP

> Use this whenever debugging an Android DAT app with local DAT Inspector MCP tools, live device events, Meta AI app/device boundary issues, permissions, registration, sessions, streaming, or user reports that the app cannot communicate with glasses.

- Skill: `facebook/live-debugging-mcp` (Agent Skill)
- Install (CLI): `npx skillmds@latest add facebook/live-debugging-mcp`
- Raw SKILL.md: https://api.skillmd.com/api/skills/facebook/live-debugging-mcp/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: facebook (https://skillmd.com/u/facebook)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/facebook/live-debugging-mcp

---


# Live DAT Debugging with MCP (Android)

Use this skill when a developer is debugging an Android DAT app and has a local
DAT Inspector MCP server or DAT debug server available to the agent. This is
the live-runtime counterpart to `dat-docs-mcp`: use docs search for API lookup,
and use this skill for observed app/device behavior.

## Ground rules

- Stay read-only unless the user explicitly asks for an app code change.
- Do not mutate Meta AI app, device, account, permission, registration, or app
  state.
- Treat companion-app/device results as boundary diagnosis from app-visible DAT
  events, not as direct access to companion app internals.
- Use live DAT evidence before guessing: registration, permissions, device
  selection, link state, session state, stream state, typed errors, and SDK logs.
- If the MCP tools are not configured, ask the developer to enable their local
  DAT debug setup and fall back to logs plus `search_dat_docs`.

## Normal agent loop

1. Discover and connect to the running app debug server:
   - `discover_debug_servers`
   - `connect_to_debug_server`
2. Establish the baseline with all three calls before diagnosing deeper:
   - `get_connection_status`
   - `get_sdk_state`
   - `get_dat_readiness`
   Do not skip `get_connection_status`; it separates MCP/debug-server
   connectivity from SDK, registration, permission, session, and stream state.
3. If the issue looks like Meta AI app or device boundary behavior, inspect:
   - `get_companion_boundary_diagnosis`
   - `get_device_path`
   - `get_permissions`
   - `get_device_properties`
4. Ask the developer to reproduce the failing Android flow while connected.
5. Wait for narrow evidence instead of polling everything:
   - `wait_for_events` with `category` or `source`
   - `get_errors`
   - `get_event_digest`
6. For handoff, collect a redacted bundle:
   - `export_diagnostic_bundle`

## Evidence map

- Initialization: app logs around `Wearables.initialize(context)`
- Registration: `Wearables.registrationState` and `wearables.registrationState`
- Device availability: `Wearables.devices` plus debug events from
  `wearables.devices`, `device.compatibility`, and `device.properties`
- Device link: `device.linkState`
- Permissions: `Wearables.checkPermissionStatus(...)`, `device.permission`,
  `check_permissions`, and `get_permissions`
- Session: `createSession(...)`, `session.start()`, `deviceSession.state`,
  `session.state`, and session errors
- Stream: `session.addCamera(...)`, `camera.stream.start()`, `camera.stream.state`, frame
  counters, and stream errors
- DAM/DWA-visible transport: `sdk.usesDam` and DAM/DWA error values

## Diagnosis patterns

- Registration blocked: verify Android manifest metadata, app credentials or
  Developer Mode setup, and callback/deeplink handling.
- No eligible device: inspect `Wearables.devices`, device compatibility, and
  `device.linkState` before changing selector code.
- Permission failure: use `get_permissions` and `get_companion_boundary_diagnosis`;
  do not assume Android runtime permission is the same as DAT permission through
  the Meta AI app/device boundary.
- Session fails before stream creation: inspect session errors and selected
  device compatibility before changing camera code.
- Stream fails after session start: inspect stream state, camera permission, and
  recent errors before changing frame processing.

## Output expectations

When summarizing, name the evidence plane, the blocking state, and the next
developer action. Example:

```text
DAT readiness is blocked at permissions: camera is denied in app-visible DAT
events. Grant the DAT camera permission in Meta AI, return to the app, then
retry session.start() and camera.stream.start() while the MCP connection waits for
permission and stream events.
```