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.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.