Skip to main content
MCP Documentation

Local Setup Guide

Install the Complete AI Assistant stack with one verified command and confirm the local model is functioning before you move on.

One-command installation

Run `curl -sSL install.completeassistant.ai | bash` to pull containers, configure environment variables, and download the compact local language model.

  • Preflight checks confirm disk space, ports, and CPU/GPU availability.
  • Installer never transmits local data; outbound network is limited to package downloads.
  • Artifacts are signed and checksummed so you can verify integrity offline.

Verification steps

When setup completes we automatically run the smoke test harness to ensure the local model produces a non-echo response and that tool catalogs are reachable.

  • `mcp doctor` validates background services, database migrations, and policy configuration.
  • Evaluation harness confirms tool allowlists and policy modes are loaded.
  • All results stream into the episodic log for future audits.

Troubleshooting

If verification fails, inspect the detailed log at `~/.mcp/logs/install.log` and review our GPU, SELinux, and WSL2 remediation steps.

  • Override default model choice with `MCP_MODEL=medium` when you have more local compute.
  • Use `--offline` to skip remote checksums when running in isolated networks.
  • Contact support through `/contact?context=setup` with the install log hash.