Config Studio — edit the config from the dashboard

The observability dashboard can show the effective config; Config Studio lets you edit it safely from the browser — with syntax highlighting, schema-driven autocomplete, live validation, per-device connection tests, a diff preview, and a safe-apply that auto-reverts if it locks you out.

Open the Configuration panel and click ✎ Edit config.

What it doesCopy Anchor Link

• Schema-aware editor — a dependency-free JSON editor with live syntax highlighting and Ctrl-Space autocomplete. Completions (keys, enum values like transport/mode) come from the live JSON Schema (GET /api/config-schema, computed from the Zod config schema), so the hints can never drift from the real validator.
• Authoritative validation — edits are validated against the same Zod schema the server boots from (POST /api/config/validate); errors are listed with their exact path.
• Secret-safe round-trip — the browser only ever sees redacted secrets («redacted»). On save, untouched sentinels are restored from the real config server-side, so you can edit freely without exposing or losing a secret. Type a new value only to change one.
• Test connection — each device entry can be probed (probeDevice) against the unsaved config before you commit, so bad credentials are caught early.
• Diff preview — see a unified diff of current vs. edited config before saving.
• Safe-apply with auto-rollback — saving writes the file (with a timestamped backup), hot-swaps the in-memory config, and starts a rollback countdown. You must click Keep changes within the window; if you don't (you locked yourself out, or the browser lost the server), the change auto-reverts to the backup — RouterOS-Safe-Mode-style, for the config file itself.

Where it writesCopy Anchor Link

Config Studio writes to the --config file it was started with, or to ~/.mikrotik-mcp/config.json when the config was assembled from env/flags. The Docker image defaults to /home/bun/.mikrotik-mcp/devices.json; a writable mount retains the complete settings there across container recreation. Normal files and directory mounts use atomic replacement. A single-file bind mount uses a flushed backup plus an in-place write when rename returns EBUSY; that fallback is not crash-atomic. See Docker persistence.

Honest constraintCopy Anchor Link

Editing an existing device's connection details is fully live (handlers read the device config per call). But the MCP tool device enum is baked at tool-registration time, so adding or removing a device name only reaches the model's tool surface after the MCP client reconnects (HTTP) or the server restarts (stdio). The panel says so when the device list changes — no fake promises.

API (token-gated, like the rest of the dashboard)Copy Anchor Link

GET  /api/config-schema           the live JSON Schema (drives autocomplete)
POST /api/config/validate         { ok, errors[] }
POST /api/config/test-device      probe a device entry before saving
POST /api/config/preview          unified diff vs current
POST /api/config                  validate → merge secrets → apply (armed rollback)
POST /api/config/keep             confirm a pending apply
POST /api/config/rollback         revert a pending apply now