Skip to main content
Glama
Luissalet

Baird's Hoard MCP Server

by Luissalet

Baird's Hoard

Live view, screenshots, instant clips and recordings of another computer plugged into this PC through an HDMI capture device, with files you can drag or paste straight into a chat, and an MCP bridge so assistants can look at that screen too.

Part of the Hoard family (Hoard Hub, Faustus). Runs locally on Windows, port 5232.

What it does

Live view

The other computer's screen in a window, 1080p, with sound, under a second of delay. The device's own H.264/AAC is copied into fragmented MP4 and played by the browser's hardware decoder: no re-encoding, almost no CPU.

Screenshot (C)

The exact current frame as PNG (or JPG), saved in a tenth of a second.

Last seconds (B)

The last 30 s (configurable, up to the memory window) saved as MP4 instantly: the app keeps the recent past in memory, cut on key frames.

Record (R)

Picture and sound until you stop; starts from the key frame before the click.

Recent files

Every file appears at once in the side list: drag the card into a chat, a mail or a folder (inside the Hoard window this is a real file drag), or press Copy and paste with Ctrl+V. Explorer shows it, Light makes a 720p copy for chats with upload limits, Delete sends it to the Recycle Bin.

Assistants

16 MCP tools. screen_look returns the current picture as an image, so an assistant can see the other computer; clip_save, record_start/stop, media_list, media_look, media_copy…

Other keys: M sound, F full screen (double click on the picture also).

Related MCP server: assistant-vision

Hardware

AVerMedia Live Gamer Portable Lite (GL310, USB id 07CA:C835); the Live Gamer Portable (C875) uses the same path. Only the device driver is needed — the WHQL-signed AVer835_x64.inf (v3.7.64.39) from AVerMedia's GL310_3.7.x.39.zip. No vendor application is used or needed. See docs/DEVICE.md for how the capture works and why ordinary capture programs cannot open this device.

Run

From Hoard Hub: the Baird's Hoard tile (start, stop, open in a window or a browser tab).

By hand:

python -m venv .venv
.venv\Scripts\python -m pip install -r requirements.txt
.venv\Scripts\python -m baird_hoard

Needs ffmpeg on PATH (or BAIRD_FFMPEG). The native helper native/bairdcap.exe is compiled on first start with the C# compiler that ships with Windows (.NET Framework 4.8).

Files go to Videos\Baird by default (Settings → Folder). Settings, thumbnails and logs live in data/.

Settings

Capture on/off (off frees the device for other programs), device, audio input (HDMI or line in), screenshot format, clip length, memory window (10–600 s), media folder, size and bitrate of light copies.

For assistants

mcp_server.py is a stdio MCP bridge (it starts the app when it is not running). Tools: capture_status, screen_look, snapshot_save, clip_save, record_start, record_stop, media_list, media_look, media_light_copy, media_copy, media_reveal, media_rename, media_delete, jobs_list, settings_get, settings_set. The same handlers serve the web UI (POST /api/ui/call) and POST /api/agent/call (Bearer token from data/mcp-token).

HTTP API

Route

GET /api/health

service, capture state, input resolution

GET /api/status

capture, recording, jobs, viewers

POST /api/snapshot, /api/clip, /api/record/start, /api/record/stop

the three buttons

GET /api/media, GET /media/{name}, GET /api/thumb/{name}

library, files, thumbnails

POST /api/media/{name}/copy · /reveal · /light · /rename, DELETE /api/media/{name}

file actions

GET/PUT /api/settings, GET /api/devices

settings

WS /api/live

live fragmented MP4 (init JSON, init segment, fragments; reset starts a new stream)

Tests

.venv\Scripts\python -m pytest

BAIRD_FAKE=1 replaces the device with an ffmpeg test pattern (used by the tests and for demos).

License

MIT. Third-party parts: THIRD_PARTY_NOTICES.md.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables AI assistants to capture screenshots and control desktop input (mouse, keyboard) to see and interact with your screen. Features user-first safety controls including automatic pause on user activity and app allowlists to restrict interactions to approved applications only.
    1
    MIT
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables AI assistants to perceive and operate macOS applications through screenshots, on-device OCR, accessibility elements, and consent-gated input, and to run an app in an isolated virtual-display session with screenshots, OCR/ASCII frames, bounded history, and evidence reports. It supports text-only models via spatial text representations while keeping capture read-only by default and gating control behind macOS permissions and operator authorization.
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables Codex to observe a physically separate computer or tablet through HDMI capture and control it with USB HID keyboard and mouse, pairing each action with the latest observation.
    2
    AGPL 3.0