Docker is the only channel for this product. GroupDocs.Parser’s engine embeds around 234 MB of ONNX models, which puts the packed tool over nuget.org’s 250 MB package limit — so dnx and dotnet tool install are not available for this product. When a slimmer engine ships, the NuGet channel returns. That also means there are no libgdiplus-style native dependencies to install — everything is inside the image.
Prerequisites
setup/linux.sh automates Debian/Ubuntu, idempotently; add --check to report without installing.
PowerShell 7 (pwsh) — runs the installer scripts.
Docker engine (sudo apt-get install -y docker.io); the script adds you to the docker group — log out/in (or newgrp docker) before the first run.
Manual run
docker run --rm -i -v $(pwd)/documents:/data \
ghcr.io/groupdocs-parser/parser-net-mcp:latest
Client config locations on Linux: Claude Desktop ~/.config/Claude/claude_desktop_config.json, VS Code user-level ~/.config/Code/User/mcp.json — full list in Register in AI clients.
Verify
pwsh ./verify-groupdocs-mcp.ps1
A passing run completes the MCP handshake (seven tools listed) and — when a document sits in your storage folder — a real get_document_info call. For headless servers, -EmitCompose generates a docker-compose.yml for supervised operation.
Linux-specific troubleshooting
docker: permission denied — your user is not in the docker group yet; newgrp docker or re-login.
Empty /data in the container — the volume path did not exist on the host, so Docker created an empty directory. Create the folder first, then run.
SELinux hosts — add :z to the volume (-v $(pwd)/documents:/data:z) if the container cannot read mounted files.
Slow first start — the image is large (it carries the ONNX models); pull it once with docker pull ghcr.io/groupdocs-parser/parser-net-mcp:latest before the first tool call.