Dashboard Guide

Web Dashboard: Global Command Center + Repo Workspace

The MCP Local Memory Dashboard serves two jobs at once: global orchestration for the agent coordinator and per-repository execution for detailed work inside one repo. It stays aligned with the same coordination model exposed by the MCP tools — tasks, claims, handoffs, standards, and memories all share one operational flow.

Key Capabilities

• Global Command Center: the Dashboard tab shows cross-repository workload, coordination pressure, throughput, and the repos that need attention first.
• Selected Repo Pulse: the same Dashboard tab keeps repo-specific memory, task, and execution metrics for the currently selected repository.
• Task Coordination Visibility: task board cards and task detail drawers surface active claims and pending handoffs directly.
• Claims Operations: the Handoffs tab shows active claims and lets you release stale ownership without leaving the dashboard.
• Handoff Operations: handoff rows expose richer transfer context including linked task_code, updated_at, expires_at, and structured context.
• Queue Operations: the Queue tab shows the embedding/KG outbox worker — status counts (pending/claimed/done/failed) plus lifetime processed/failed/poisoned counters — and a Failed-jobs table with per-row Re-run and Clear actions (plus Re-run all).
• Reference Alignment: dashboard actions and MCP tools follow the same flow for task status updates and coordination cleanup.

How to Start

Run the dashboard from the repository root:

npx @vheins/local-memory-mcp dashboard

Then open http://localhost:3456.

For a local source checkout, the dashboard bin automatically rebuilds the served UI bundle (dist/dashboard/public) when src/dashboard/ui is newer — so editing the dashboard sources and re-running the command serves the fresh build automatically. To build explicitly, run npm run dashboard:build before serving. (In an installed npm package the src/ directory is not shipped, so the published bundle is served as-is.)

How to Use

1. Orchestrator Mode

Use the Dashboard tab before drilling into a repo.

• Review Global Command Center for repo count, active repos, blocked tasks, active claims, and pending handoffs.
• Use the Attention Board to identify which repositories have the highest pressure from blocked work, queue buildup, or coordination overhead.
• Keep one repository selected in the sidebar to maintain a live repo pulse underneath the global overview.

2. Repository Execution Mode

Once a repo is selected:

• Open the Tasks tab for the kanban board.
• Watch task cards for active coordination badges showing current claim owners or pending handoffs.
• Open the task drawer to inspect detailed coordination state before moving a task or taking ownership.

3. Claims and Handoffs

Use the Handoffs tab as the coordination console.

• Claims: inspect active ownership across the selected repo and release stale claims when reassignment is needed.
• Handoffs: create transfer context for unfinished work, inspect structured context, and close consumed or stale handoffs.
• Cleanup: completing or canceling a task through the dashboard follows the MCP task-write status rules, so active claims are released and linked pending handoffs are expired automatically.

Coordination Model

The dashboard mirrors the MCP tool flow:

• claim-manage creates ownership (claim mode)
• claim-manage inspects ownership (list mode)
• claim-manage clears stale ownership (release mode)
• handoff-write transfers unfinished work
• handoff-read inspects queue state
• handoff-write closes stale or consumed transfers (update mode)
• task-write is the authoritative status transition path

This means the dashboard is no longer a separate mutation path for tasks. Coordination cleanup and task lifecycle rules stay consistent between UI and MCP callers.

Queue Tab

The Queue tab (the 11th tab, after Handoffs and before Knowledge Graph) surfaces the embedding/KG outbox worker and its failed jobs.

Server-wide by design: the queue is a shared outbox for the whole server, not per-repo. When no repository is selected, the Queue tab shows jobs from all repos (each job row is tagged with its entity's repo). Select a repository in the sidebar to filter the view (?repo=).

• Status summary: counts by state (pending/claimed/done/failed) with lifetime processed/failed/poisoned counters and a worker running indicator — from GET /api/queue/status
• Failed jobs table: terminal poison jobs with entity kind/id, attempts/max attempts, enqueued/processed timestamps, and the last error
• Re-run: re-queue a failed job so the worker processes it again (per row, or all at once)
• Clear: delete a failed job row (asks for confirmation)

Failed jobs are named poison in the queue API and rendered as "Failed" in the UI.

Codebase Tab

The Codebase tab (between Tasks and Handoffs) browses the codebase index of the selected repo:

• Symbol search + file tree — search symbols or browse the recursive file tree; select a file to view its indexed content in the file viewer
• Symbol detail — click a symbol to inspect its definition, references, and trace; the detail includes a call-graph DAG (callers/callees)
• Index stats — status/statistics panel plus language and symbol-kind breakdown
• Code-graph panel — a force-directed code graph of the repo (kind filter All/Calls/Imports/Co-defined, zoom/refresh, legend footer) fed by GET /api/codebase/graph; clicking a node opens its symbol detail. If the repo is not indexed it shows "Index required — run codebase-index"

Results are typically fresh: a polling watcher (default on) re-indexes changed files within ~30s (FILE_WATCH_INTERVAL_MS), so edits usually appear without an explicit re-index. See the Codebase Index feature page.

Knowledge Graph Tab

The Knowledge Graph tab provides an interactive force-directed graph visualization of entities and their relations.

• Color-coded nodes by entity type: green (person), blue (place), orange (organization), purple (concept)
• Edge labels showing relation types between connected entities
• Click a node to inspect its details and observations in a tooltip
• Double-click a node to delete it (asks for confirmation; cascades to relations and observations)
• Right-click an edge to delete the relation
• Add Entity modal: name, type dropdown (person/place/organization/concept), description
• Add Relation modal: from-entity select, to-entity select, relation type input

Use this tab to visually map your domain knowledge, discover hidden connections, and manage the Knowledge Graph without writing MCP tool calls.

Notes

• Most tabs remain per-repository.
• The Dashboard tab is intentionally hybrid: global overview first, selected repository pulse second.
• If no repository is selected, the Dashboard tab still works in global mode.

Disclaimer

THE DASHBOARD IS PROVIDED "AS IS", without warranty of any kind. It is intended for manual inspection and management of local data.

Related Pages

• Tool Reference & Usage Guide — the underlying MCP tools
• Codebase Index — symbol indexing and search
• Auto-Start Dashboard — launch the dashboard with your IDE
• Troubleshooting — dashboard issues and fixes
• Panduan Dasbor Web — Bahasa Indonesia