Step 1
Configure provider credentials safely
Keep model provider secrets in your own secret manager and reference them from BlackShield configuration.
- Create provider credentials for your approved model endpoint.
- Map credential references through `AI_SHIM_SECRET_VALUES` using `credentials_ref` keys.
- Enable only the AI tasks your team intends to use.
Step 2
Enable AI-assisted workflows with guardrails
Start with high-value workflows and enforce human review for production decisions.
- Enable `smart_alert_triage` for backlog prioritization.
- Enable `pr_diff_analysis` for secure code review support.
- Set model, token, timeout, and temperature limits aligned to your risk policy.
Step 3
Integrate your MCP client
Configure your MCP client to call BlackShield tools while keeping model inference under your control.
- Set MCP endpoint to `/api/v1/mcp` and include workspace JWT in the Authorization header.
- Run `initialize` and `tools/list` before calling `tools/call`.
- Use `/api/v1/mcp/manifest` to discover available tool metadata.
- Keep provider keys in your own runtime and require human approval for production changes.