Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@agentic-mcp-server show active agents and recent tool calls"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agentic-mcp-server · Real-time observability for multi-agent development workflows
Mission Control is a real-time dashboard for monitoring AI agents connected to the agentic-mcp-server. It shows which agents are active, what features are being worked on, every tool call as it happens, and aggregated usage statistics — all updating live via WebSocket.
Live URL: https://agentic-dashboard.365softlabs.com
Worker API: https://agentic-mcp-server.{account}.workers.dev
Related MCP server: cco-mcp
┌─────────────────────────────────────────────────────────────────────┐
│ AI AGENTS (Clients) │
│ │
│ Claude Desktop Claude Code Roo Code CodeGPT │
│ role=planner role=frontend role=tester role=backend │
└────────────────────────────┬────────────────────────────────────────┘
│ MCP over HTTP/SSE
│ POST /mcp?role=X&name=Y
▼
┌─────────────────────────────────────────────────────────────────────┐
│ CLOUDFLARE WORKER │
│ agentic-mcp-server │
│ │
│ ┌──────────────────────────────┐ ┌────────────────────────────┐ │
│ │ AgenticMcpAgent DO │ │ DashboardBroadcaster DO │ │
│ │ (one per agent session) │ │ (one global instance) │ │
│ │ │ │ │ │
│ │ • Handles MCP tool calls │ │ • Holds all WebSocket │ │
│ │ • Wraps every tool in │ │ connections from │ │
│ │ callTool() logger │ │ dashboard clients │ │
│ │ • Persists session state │ │ • Fans out events to │ │
│ │ in Durable Object storage │ │ all connected dashboards│ │
│ │ • Calls logActivity() + │ │ │ │
│ │ broadcast() after │ │ POST /broadcast │ │
│ │ every tool call │ │ GET /ws (WebSocket) │ │
│ └──────────┬───────────────────┘ └─────────────▲──────────────┘ │
│ │ │ │
│ │ fire-and-forget │ broadcast event │
│ ▼ │ │
│ ┌──────────────────────────────┐ │ │
│ │ Cloudflare KV │ │ │
│ │ SHARED_CONTEXT │ │ │
│ │ │ │ │
│ │ dashboard:sessions │ │ │
│ │ dashboard:activity:log │ │ │
│ │ dashboard:stats │ │ │
│ │ feature:{id} │ │ │
│ │ feature:index │ │ │
│ │ conventions:global │ │ │
│ │ memory:{role} │ │ │
│ │ notes:shared │ │ │
│ └──────────────────────────────┘ │ │
│ │ │
│ Dashboard API Routes: │ │
│ GET /dashboard/api/sessions ─────────────────►│ │
│ GET /dashboard/api/features ─────────────────►│ │
│ GET /dashboard/api/activity ─────────────────►│ │
│ GET /dashboard/api/stats ─────────────────►│ │
│ GET /dashboard/ws (WebSocket) ──────────────────┘ │
└─────────────────────────────────────────────────────────────────────┘
│
│ REST API + WebSocket
▼
┌─────────────────────────────────────────────────────────────────────┐
│ CLOUDFLARE PAGES │
│ agentic-mcp-dashboard │
│ │
│ React SPA (Vite build) │
│ https://agentic-dashboard.365softlabs.com │
│ │
│ • Initial load: fetches all 4 API endpoints in parallel │
│ • Real-time: WebSocket receives events and updates UI instantly │
│ • Fallback: polls all endpoints every 30 seconds │
└─────────────────────────────────────────────────────────────────────┘When an AI agent connects to /mcp?role=frontend&name=claude-code:
A new AgenticMcpAgent Durable Object is created (or existing one resumed)
init() runs — restores session state from DO storage
registerSession() writes the session to dashboard:sessions in KV
Dashboard receives no real-time event for this (sessions are polled)
Every tool call (e.g. get_feature_spec, update_task_status) is wrapped in callTool():
Agent → tool call → callTool() wrapper
│
├─► executes the actual tool fn
│
└─► in finally block (fire-and-forget):
├─► logActivity() → writes to KV
│ dashboard:activity:log (last 200 events)
│ dashboard:stats (aggregated counters)
│
├─► broadcast() → POST to DashboardBroadcaster DO
│ fans out to all WebSocket clients
│
└─► touchSession() → updates session lastActiveAt in KVThe WebSocket client in the dashboard (hooks.ts) receives the broadcast and immediately updates React state — no re-fetch needed:
Event Type | What Updates |
| Activity feed prepended, session |
| Kanban board card updated or new card added |
| Task chip status updated within feature card |
| No UI update (logged only) |
All endpoints are served by the Worker at /dashboard/api/*.
GET /dashboard/api/sessionsReturns the session registry — all known agent sessions.
{
"sessions": {
"session-id-xyz": {
"sessionId": "session-id-xyz",
"agentRole": "frontend",
"agentName": "claude-code",
"createdAt": "2026-02-28T10:00:00Z",
"lastActiveAt": "2026-02-28T11:30:00Z",
"currentFeature": "feat-001",
"currentTask": "feat-001-task-2",
"isActive": true
}
}
}Sessions inactive for more than 2 hours are automatically marked isActive: false.
GET /dashboard/api/featuresReturns all feature specs stored in KV.
[
{
"id": "feat-001",
"title": "User Authentication",
"description": "OAuth2 login via GitHub",
"status": "in-progress",
"createdAt": "2026-02-28T09:00:00Z",
"updatedAt": "2026-02-28T11:00:00Z",
"tasks": [
{
"id": "feat-001-task-1",
"title": "Design auth flow",
"assignedRole": "planner",
"status": "done",
"completedAt": "2026-02-28T10:00:00Z"
},
{
"id": "feat-001-task-2",
"title": "Implement Worker endpoint",
"assignedRole": "frontend",
"status": "in-progress"
}
],
"assignedAgents": {
"planner": "claude-desktop",
"frontend": "claude-code"
},
"githubBranch": "feat/user-authentication"
}
]GET /dashboard/api/activity?limit=100Returns the most recent activity events (default 50, max 200).
[
{
"id": "evt-1709118000000-abc12",
"timestamp": "2026-02-28T11:30:00Z",
"sessionId": "session-id-xyz",
"agentRole": "frontend",
"agentName": "claude-code",
"tool": "update_task_status",
"featureId": "feat-001",
"taskId": "feat-001-task-2",
"durationMs": 142,
"success": true
}
]GET /dashboard/api/statsReturns aggregated usage statistics.
{
"lastUpdated": "2026-02-28T11:30:00Z",
"totalToolCalls": 847,
"toolCallCounts": {
"get_feature_spec": 312,
"update_task_status": 201,
"write_shared_memory": 98
},
"agentCallCounts": {
"claude-code": 445,
"codegpt": 312,
"roo-code": 90
},
"roleCallCounts": {
"frontend": 445,
"backend": 312,
"tester": 90,
"planner": 0,
"reviewer": 0,
"orchestrator": 0
},
"featureActivityCounts": {
"feat-001": 523,
"feat-002": 324
},
"dailyCounts": {
"2026-02-28": 847
},
"hourlyActivity": [0,0,0,0,0,0,0,12,45,98,120,87,65,43,21,18,32,54,76,43,21,0,0,0]
}GET /dashboard/wsWebSocket endpoint. Upgrades to WebSocket and subscribes to real-time broadcast events from DashboardBroadcaster.
Events pushed to all connected dashboard clients in real time:
tool_callFired after every MCP tool invocation.
{
"type": "tool_call",
"timestamp": "2026-02-28T11:30:00Z",
"sessionId": "session-id-xyz",
"agentRole": "frontend",
"agentName": "claude-code",
"tool": "update_task_status",
"featureId": "feat-001",
"taskId": "feat-001-task-2",
"durationMs": 142,
"success": true
}feature_updatedFired when upsert_feature_spec is called.
{
"type": "feature_updated",
"timestamp": "2026-02-28T11:30:00Z",
"feature": { ...full FeatureSpec object... }
}task_updatedFired when update_task_status is called.
{
"type": "task_updated",
"timestamp": "2026-02-28T11:30:00Z",
"featureId": "feat-001",
"taskId": "feat-001-task-2",
"status": "done",
"agentRole": "frontend",
"agentName": "claude-code"
}memory_writtenFired when write_shared_memory is called.
{
"type": "memory_written",
"timestamp": "2026-02-28T11:30:00Z",
"key": "memory:frontend",
"role": "frontend",
"agentName": "claude-code"
}Always visible across the top.
Element | Description |
MISSION CONTROL logo | App title and branding |
UTC Clock | Live clock updating every second |
Active agents | Count of sessions with |
LIVE / RECONNECTING | WebSocket connection status — green when connected, red when reconnecting |
↺ REFRESH | Manual trigger to re-fetch all API endpoints |
Six metrics displayed as a grid below the header, always visible.
Metric | Source | Description |
TOTAL CALLS |
| All-time tool invocations |
FEATURES LIVE |
| Currently active features |
IN REVIEW |
| Features awaiting review |
SHIPPED |
| Completed features |
TOP AGENT |
| Most active agent by name |
TOP TOOL |
| Most called tool |
Four columns representing the feature lifecycle:
Planning → In Progress → Review → DoneEach feature card shows:
Feature ID (e.g. feat-001) and title
Progress bar (% of tasks with status === 'done')
Task count and GitHub branch name (if set)
Expandable task list — click card to reveal individual tasks with role badges and status dots
Task status colours:
todo → dim grey
in-progress → cyan
done → green
blocked → red
List of all known agent sessions, sorted by most recently active.
Each session card shows:
Agent name with status dot (green = active, grey = inactive)
Role badge (colour-coded by role)
Current feature being worked on
Time since last activity
Role colours:
Role | Colour |
Planner | Cyan |
Frontend | Green |
Backend | Purple |
Tester | Yellow |
Reviewer | Orange |
Orchestrator | Red |
Sessions inactive for >2 hours are automatically marked inactive by the Worker and shown in grey.
Live scrolling feed of all tool calls, newest at top. Maximum 100 events shown (last 200 stored in KV).
Each entry shows:
Tool icon (emoji)
Agent name (colour-coded by role)
Tool name called
Feature ID badge (if applicable)
Time since event
Duration in milliseconds
ERR badge if success === false
New events arriving via WebSocket briefly highlight in cyan with a slide-in animation.
Three visualisations:
Hourly Activity Chart — 24-bar chart showing tool calls per UTC hour. Current hour highlighted in cyan with glow effect. Resets conceptually per hour (sliding counter).
Top Tools — horizontal bar chart of most-called tools, sorted descending.
Top Agents — horizontal bar chart of most active agents by name.
Top Features — horizontal bar chart of most active features by ID.
All data is stored in the SHARED_CONTEXT KV namespace.
Key | Type | Description |
|
| Coding standards, read by all agents on session start |
|
| List of all feature IDs |
|
| Full feature spec including tasks |
|
| Role-specific shared memory (e.g. |
|
| General shared notes visible to all agents |
|
| All agent session metadata |
|
| Last 200 tool call events |
|
| Aggregated counters and hourly chart data |
Add to claude_desktop_config.json:
{
"mcpServers": {
"agentic-mcp": {
"command": "npx",
"args": [
"mcp-remote",
"https://agentic-mcp-server.{account}.workers.dev/mcp?role=planner&name=claude-desktop"
]
}
}
}claude mcp add agentic-mcp \
--url https://agentic-mcp-server.{account}.workers.dev/mcp \
--params "role=frontend&name=claude-code"In .roo/mcp.json:
{
"mcpServers": {
"agentic-mcp": {
"url": "https://agentic-mcp-server.{account}.workers.dev/mcp",
"params": { "role": "tester", "name": "roo-code" }
}
}
}In CodeGPT settings → MCP Servers → Add Remote:
URL: https://agentic-mcp-server.{account}.workers.dev/mcp?role=backend&name=codegpt
Transport: Streamable HTTPOnce connected, the agent should call get_conventions first — this will immediately appear in the Activity feed and increment the stats counters.
npx wrangler deploy# From dashboard/ directory
$env:VITE_API_URL="https://agentic-mcp-server.{account}.workers.dev"; npm run build
npx wrangler pages deploy dist --project-name=agentic-mcp-dashboardVariable | Value |
|
|
|
|
|
|
Secret | Purpose |
| 32-char random string for session token signing |
| GitHub OAuth app client ID |
| GitHub OAuth app client secret |
This server cannot be deployed
AI agent observability for production traces, natural-language insights, and improvement loops.
Data + AI observability — monitor and troubleshoot production-grade agents and the context they use.
See, price, and control every tool call your AI agents make: policy checks, cost, and audit tools.
Production observability for AI agents: search runs, read evaluations, acknowledge incidents.
