Local Snagit MCP Server
Lets GitHub Copilot record demo videos using locally installed TechSmith Snagit, capturing displays, windows, or regions to MP4 and controlling the recording flow through MCP tools.
README
Local Snagit MCP Server
A local MCP server that lets GitHub Copilot record demo videos with the copy of TechSmith Snagit already installed on your machine. Copilot drives the recording while it drives everything else — a terminal walkthrough, a deployment, or a Playwright session against your deployed app — and gets back an MP4.
It talks to Snagit through Snagit's own SNAGIT.VideoCapture COM automation interface, so there
are no hotkey simulations, no UI scripting, and no third-party screen recorder.
Requirements
- Windows
- Snagit 2023 or newer, installed and launched at least once (verified against Snagit 2025, 25.4.0.8498)
- Python 3.10+
Setup
py -3.12 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e .
.\.venv\Scripts\python.exe -m snagit_mcp --selftest
--selftest should report "ready": true. If it does not, see Troubleshooting.
The repo ships .vscode/mcp.json, so VS Code offers to start the server as soon
as you open the folder. To use it from any workspace, add the same block to your user-level
mcp.json with absolute paths instead of ${workspaceFolder}.
Recordings land in SNAGIT_MCP_OUTPUT_DIR (the workspace recordings/ folder by default, or
%USERPROFILE%\Videos\Snagit MCP when unset).
| Environment variable | Default | Purpose |
|---|---|---|
SNAGIT_MCP_OUTPUT_DIR |
%USERPROFILE%\Videos\Snagit MCP |
Where MP4s are written |
SNAGIT_MCP_MAX_WAIT_SECONDS |
600 |
Upper bound for the wait tool |
SNAGIT_MCP_ENCODE_TIMEOUT_SECONDS |
300 |
How long stop_recording waits for encoding |
Tools
| Tool | What it does |
|---|---|
check_snagit_setup |
Snagit version, COM reachability, output folder, display count |
list_displays |
Monitors and their pixel bounds |
list_windows |
Visible windows with handles and bounds — use it to find the browser |
focus_window |
Restore and foreground a window before recording it |
start_recording |
Record a display, a window, or a fixed region to MP4 |
recording_status |
Whether a recording is running, and for how long |
pause_recording / resume_recording |
Skip dead air such as a deployment finishing |
stop_recording |
Stop, wait for encoding, return the MP4 path and duration |
wait |
Hold a beat so the last on-screen action is readable |
list_recordings |
Most recent MP4s in the output folder |
There is also a record_app_demo prompt (a slash command in VS Code) that gives Copilot the full
shot-list-first procedure.
Recording a demo of a deployed app
Snagit records real pixels, so the browser has to be visible — run Playwright headed, not headless. Combined with the Playwright MCP server, a session looks like this:
- Playwright MCP opens your App Service URL in a headed browser and signs in.
list_windowswithtitle_contains: "Edge"(orChrome) returns the browser window handle.start_recordingwithtarget: "window"and that handle. It returns after Snagit's countdown, so the next action is already on camera.- Playwright clicks through the scenario;
waitbetween steps keeps it followable. stop_recordingreturns the MP4 path.
For a repo walkthrough that moves between VS Code, a terminal, the Azure portal, and the browser,
use target: "fullscreen" with the display_index from list_displays instead of a single window.
Ask for it in plain language, for example:
Record a demo of the deployment flow: show the prerequisites in the README, run the deploy script, then open the App Service URL and walk through creating a record. Use display 0.
How it drives Snagit
These behaviours were established by experiment and the timings in src/snagit_mcp/recorder.py depend on them:
- The sequence is
Capture()to arm, a 3s settle,Start(), thenStop(). CallingStart()too soon afterCapture()produces an empty 2 KB MP4. - Snagit plays a ~3.4s countdown after
Start().start_recordingabsorbs it and only returns once real recording has begun. - Snagit's recording toolbar must stay visible. Setting
HideRecordingUI = Truesilently truncates recordings to a couple of seconds. The toolbar sits outside the captured region, so it does not show up in the video. - Microphone and system audio are not settable over COM in this Snagit build — those setters throw, and a failed write can wedge the capture engine. Audio therefore follows whatever is configured in Snagit's own capture settings. Set it there before recording if you want narration.
RecordingDuration,FrameCountandAverageFrameRateare only meaningful afterStop(); during a recordingrecording_statusreports the server's own wall-clock instead.- Output goes straight to a file. Snagit Editor does not pop up between takes.
Verifying after a Snagit upgrade
.\.venv\Scripts\python.exe tests\smoke_recording.py
Records a real 5-second clip through the MCP tools and checks the resulting MP4's duration. If a future Snagit release changes the countdown or arming behaviour, this is what will catch it.
Troubleshooting
| Symptom | Fix |
|---|---|
check_snagit_setup reports a COM error |
Launch Snagit once so it registers its COM server, then retry |
| Recordings are empty or a couple of seconds long | The capture engine is wedged: Stop-Process -Name SnagitCapture -Force, then relaunch SnagitCapture.exe |
Snagit did not arm the recorder |
A capture is already open in Snagit — cancel it and retry |
| Region coordinates look shifted on a scaled display | The server sets per-monitor DPI awareness at startup; confirm dpi_awareness in check_snagit_setup |
A note on what ends up on camera
Screen recording captures whatever is on screen, including terminal output, tokens, portal sessions, and notifications. Close or clear anything sensitive before recording, and review the MP4 before sharing it.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。