# Architecture decision record

## ADR-0001 — Begin with a read-only integration

**Status:** Accepted  
**Date:** 2026-08-24

### Decision

The first milestone will expose only device inventory, status, and safe metadata. Remote command execution and state-changing actions will not be part of the initial MVP.

### Rationale

This validates connectivity, authentication, data mapping, logging, and OpenClaw tool behavior while minimizing the blast radius of implementation mistakes.

### Consequences

- Initial usefulness is limited to visibility and reporting.
- Diagnostics require a later, separately reviewed capability.
- The bridge can be tested without risking endpoint changes.

## ADR-0002 — Prefer a local bridge over browser automation

**Status:** Proposed  
**Date:** 2026-08-24

### Decision

Investigate a local API/MCP-style bridge first. Browser automation remains a fallback for discovery or occasional operations, not the primary architecture.

### Rationale

A typed local interface is easier to constrain, test, audit, and maintain than UI selectors and browser sessions.

### Open questions

- What supported MeshCentral API surface is available in the deployed version?
- Should the bridge run in the MeshCentral LXC, a separate LXC, or on ledclaw01?
- Which OpenClaw integration mechanism best fits the final tool contract?