AI Agent Hub

The AI Agent Hub is the full-stack dashboard for the MCP AI Suite. It wires the kernelmcp orchestrator to the six suite pipelines (memory, planning, workspace, sandbox, scheduler, RAG) behind a single FastAPI backend, with a Next.js 14 frontend for chat, agents, knowledge, observability, scheduling, security, and more.

It is a deployable application, not a pip install library — you run it with Docker (or locally) rather than importing it.

Stack

Backend — FastAPI + kernelmcp orchestrator + LiteLLM
Frontend — Next.js 14 (App Router) + Tailwind CSS + TypeScript
Search — SearXNG (meta search engine for the web-search tools)
Packaging — Docker Compose

Install & run

Docker (recommended)

git clone https://github.com/gashel01/kernelmcp-demo
cd kernelmcp-demo
docker-compose up --build

Then open:

Service	URL
Frontend (dashboard)	http://localhost:3007
Backend API	http://localhost:8007
API docs (Swagger)	http://localhost:8007/docs
SearXNG	http://localhost:9999

Local (without Docker)

Backend

cd backend
pip install -r requirements.txt
pip install mcpaisuite-kernelmcp mcpaisuite-memorymcp mcpaisuite-ragmcp mcpaisuite-planningmcp mcpaisuite-workspacemcp mcpaisuite-sandboxmcp mcpaisuite-schedulermcp
uvicorn server:app --reload --port 8000

Frontend (new terminal)

cd frontend
npm install
NEXT_PUBLIC_API_URL=http://localhost:8000 npm run dev

Open http://localhost:3000.

Dashboard pages

Page	Path	What it does
Chat	`/chat`	Multi-turn chat with streaming and ReAct / LTP / Hybrid mode selection
Agents	`/agents`	Spawn and monitor sub-agents / TaskForce (code, research, file, memory, plan, rag, ltp, custom)
Knowledge	`/knowledge`	RAG file upload, vector search, knowledge-graph explorer
Workspace	`/workspace`	File manager with checkpoints, DLP, and approval gates
Scheduler	`/scheduler`	Cron, one-shot, interval and watch jobs with execution history
Security	`/security`	Network egress control, host-access gates, constitution editor
Observability	`/observability`	Traces, token / cost / latency analytics
Monitor	`/monitor`	Live execution event stream and tool-dispatch logs
Deployments	`/deployments`	Deploy agents as token-authed callable APIs
Fleet	`/fleet`	Multi-agent fleet management
Control	`/control`	Remote control plane for connected kernels
Eval	`/eval`	Evaluation suites and regression tracking
Executions	`/executions`	Execution and run history
Observatory	`/observatory`	Aggregate system overview
Settings	`/settings`	LLM provider, model, execution mode, service connections

Configuration

Set these as environment variables (e.g. in docker-compose.yml or your shell):

Variable	Default	Description
`KERNELMCP_NAMESPACE`	`demo`	Default tenant namespace
`KERNELMCP_MODEL`	`claude-sonnet-4-6`	Primary LLM model
`KERNELMCP_LOCAL_MODEL`	`ollama/mistral`	Local / fallback model
`KERNELMCP_ROUTING`	`true`	Enable smart model routing
`KERNELMCP_MAX_TURNS`	`20`	Max ReAct turns per task
`KERNELMCP_MAX_TOKENS`	`50000`	Max tokens per task
`KERNELMCP_MAX_COST`	`1.0`	Max cost per task (USD)
`OPENAI_API_KEY`	—	OpenAI API key
`ANTHROPIC_API_KEY`	—	Anthropic API key
`SEARXNG_URL`	`http://searxng:8080`	SearXNG URL
`RAGMCP_VECTORSTORE`	`qdrant`	Vector store backend
`RAGMCP_VECTORSTORE_URL`	`http://host.docker.internal:6333`	Qdrant URL
`RAGMCP_EMBEDDER`	`fastembed`	Embedding backend
`NEXT_PUBLIC_API_URL`	`http://localhost:8007`	API URL the frontend calls

API

The backend exposes 200+ REST endpoints (chat, agents, knowledge, scheduler, observability, deployments, fleet, eval, hub, …), all accepting an X-Tenant-ID header for multi-tenant isolation. Browse the full, always-current spec at /docs (Swagger) on the running backend, or read the source on GitHub.

License

AGPL-3.0. For commercial use in a closed-source product, a commercial license is available — contact us by email.