Server Runtime Auth and Artifact Sync

How hosted agents authenticate and consume local sister artifacts safely.

Cloud Boundary

Hosted runtimes cannot directly read files from your laptop.
Sync artifacts to server-accessible storage, then authenticate server MCP calls.

1) Enable Server Mode

Required environment:

export AGENTRA_RUNTIME_MODE=server
export AGENTIC_TOKEN="$(openssl rand -hex 32)"
export AGENTRA_ARTIFACT_DIRS="/srv/agentra:/data/brains"
# optional file-based token:
# export AGENTIC_TOKEN_FILE="/etc/agentra/token"

Run strict preflight:

agentra server preflight --strict

2) Sync Local Artifacts to Server

Push runtime artifacts before expecting server takeover:

./sync_artifacts.sh --target=<server-path-or-rsync-target>

Sync includes:

• .amem
• .avis
• .acb
• optional ~/.brain.amem

3) Start MCP Runtimes on Server Host

agentic-memory-mcp serve
agentic-vision-mcp serve
acb-mcp serve

Expected install + restart signal:

[####################################] 100% Install complete
Install complete: <runtime>
[next] Restart MCP host/client
[tip] Optional feedback: https://agentralabs.tech/docs/feedback

4) Validate from Any MCP Client

Client-side checklist:

which agentic-memory-mcp
which agentic-vision-mcp
which acb-mcp
# then trigger first MCP tool call from your client

Protocol Parity

Desktop and server follow the same MCP contract.
Server mode only adds explicit authentication and artifact-sync requirements.

Operational Notes

• Keep AGENTIC_TOKEN secret; prefer AGENTIC_TOKEN_FILE in production.
• Treat synced artifact directories as sensitive data stores.
• Re-run agentra server preflight --strict after deploys, token rotation, or path changes.