SuperAgenticMCP
This server currently exposes only two read-only stub endpoints for inspecting the router.
status — returns router status: version, racked servers (stub), and health.
list_rack — lists currently racked MCP servers (stub).
No actual task planning, routing, tool calls, or memory operations are exposed yet; the schema reflects a minimal scaffold rather than the full agent-swarm control plane described in the README.
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., "@SuperAgenticMCP Plan this task, route each step to the right MCP server, and persist the results."
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.
SuperAgenticMCP
The switchboard for your agent swarm.
SuperAgenticMCP is a production-oriented, MCP-native router and multi-agent control plane. It sits between your agents and the MCP servers they need — planning the task, routing each tool call to the right server, streaming every step, and filing results into memory.
Patch every agent into every tool.
Support Agentic OS Kernels ($99) · Agentic OS Cycle ($0.75) · Public Goods Support · USDC
Layout standard: Server OS — see docs/LAYOUT.md.
Surfaces
Surface | Entry |
Web control plane |
|
MCP Server |
|
CLI |
|
SDK |
|
Board (target) |
|
Skills |
|
AGENTS.md | Coding-agent contract |
Verify |
|
CI |
|
Related MCP server: agentloop
Why SuperAgenticMCP?
Problem | Approach |
Agents need many MCP servers | One process racks servers and routes by capability |
Tool calls are opaque | Live patch board + JSONL log + run receipt |
Multi-agent coordination is ad-hoc | Built-in planner / worker / critic loop |
Memory evaporates between runs | Memory scope constellation of artifacts |
Config drifts from reality | Rack toggles generate |
Features
Capability | Status |
MCP over stdio (+ streamable HTTP planned) | ✅ scaffold |
Capability-first routing + latency tie-break | 🚧 / demo in hero |
Planner → parallel workers → critic | 🚧 / demo in hero |
Hot-swap server rack | 🚧 / demo in hero |
Live patch board + phosphor log | ✅ hero |
Memory scope (3D constellation) | ✅ hero |
Call budgets + offline skip | 🚧 |
Dynamic | 🚧 |
Server OS layout + verify contract | ✅ |
Self-learning model routing (bandit) | 🚧 |
Agentic commerce (x402) | ✅ routes to x402-cloudflare-starter |
Commerce is not reimplemented here. Rack the x402 Worker; keep Stripe Payment Links for humans. Receipts are not Desk unlocks.
Quick start
git clone https://github.com/ANAMIZED/SuperAgenticMCP
cd SuperAgenticMCP
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
superagenticmcp # MCP server (stdio)
superagenticmcp-cli status
make test
bash scripts/verify.shZero-install hero:
MCP client config
{
"mcpServers": {
"superagenticmcp": {
"command": "superagenticmcp",
"args": []
}
}
}Docker (scaffold)
docker compose up --buildHow routing works
Protocol — Model Context Protocol; one session per racked server.
Agent loop — Planner decomposes; workers run in parallel where allowed; critic checks before ship.
Routing — Capability match first, latency as tie-break, fallback when a server drops.
Guardrails — Per-server allow-lists and per-run call budgets.
Telemetry — JSONL lines + board replay.
Footprint — Single process for the core path; no DB required.
Repository layout
src/superagenticmcp/ # MCP, CLI, SDK
skills/ # Packaged skills (Server OS style)
web/ # Control-plane path
scripts/verify.sh # Verify contract
tests/ # Smoke + contract tests
superagenticmcp.html # Hero demo
docs/LAYOUT.md # Permanent layout standard = Server OS
AGENTS.md # Agent contractSupport
Fiat (Stripe)
Offer | Link |
Support Agentic OS Kernels — $99 | |
Agentic OS Cycle — $0.75 | |
Public Goods Support |
GitHub Sponsors: ANAMIZED
USDC (crypto)
Non-custodial addresses controlled by ANAMIZED. Pure x402 micropayments preferred for agents — see x402-cloudflare-starter.
Network | Address | Explorer |
Base |
| |
Ethereum |
| |
Solana |
|
Send only USDC on the matching network. Wrong asset or chain can mean permanent loss.
Related projects
server-os — layout standard · agents as processes
OpenGOS — grants + public-goods funding MCP
LRSI — local recursive self-improvement kernel
x402-cloudflare-starter — USDC micropayments (Base + Solana)
License
Apache-2.0
Built for the Model Context Protocol · Unit 001 · SA-MCP MK.4 · Server OS layout
Available Tools
5 toolslist_rackList racked MCP serversA
List MCP servers currently mounted on the SuperAgentic rack (name, command or URL, notes). Use after rack_add to confirm membership. Empty list means no servers have been registered in this process. Read-only; use rack_add / rack_remove to mutate. Does not probe remote health.
Inputs: no named parameters. Returns: object
rack_addAdd a server to the rackA
Register an MCP server on the in-memory rack so later route_task calls can target it by name. Overwrites an existing entry with the same name (idempotent upsert). Does not start the child process and does not validate the command. Use list_rack to inspect and rack_remove to delete. Local write only.
Inputs: name (required), command (required), notes. Returns: object
rack_removeRemove a server from the rackA
Remove one MCP server from the in-memory rack by name. Destructive. Missing names return an error object rather than raising. Does not stop a running child — only drops the catalog entry. Use list_rack first to confirm the name.
Inputs: name (required). Returns: object
route_taskPlan a task against the rackA
Produce a routing plan for a natural-language task against the current rack. In this release the planner is a deterministic stub: it lists candidate servers and does not execute downstream tools. Use after rack_add. Not a substitute for status. Read-only with respect to the rack; no network calls yet.
Inputs: task (required), preferred_server. Returns: object
statusGet router statusA
Return SuperAgenticMCP router health: version, process status, and how many MCP servers are currently racked. Use this as a connectivity probe before list_rack or route_task. Read-only local state. Does not call downstream MCP servers (not open-world). Pair with list_rack for the actual catalog.
Inputs: no named parameters. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool has a distinct role: status reports router health, list_rack shows catalog contents, rack_add and rack_remove mutate the rack, and route_task plans routing. There is no meaningful overlap or ambiguity between tools.
Most tools follow a clear verb_noun snake_case pattern such as list_rack, rack_add, and rack_remove. The lone status tool breaks the verb_noun pattern, but it is still a simple, readable name and does not cause confusion.
Five tools is well-scoped for a router/rack management server. Each tool covers a necessary operation without redundancy or unnecessary bloat.
The surface covers the core lifecycle: health check, catalog listing, add/upsert, remove, and routing plan generation. Minor gaps exist such as no single-server detail endpoint and route_task being a deterministic stub, but the main workflows are supported.
Maintenance
Related MCP Connectors
- aXDeprecatedio.github.ax-platform
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Related MCP Servers
- FlicenseNot gradedqualityBmaintenanceA production-ready MCP gateway and control plane that provides credential vault, policy engine, audit logging, and managed runtime for routing tool calls between AI agents and downstream MCP servers.59-
- AlicenseNot gradedqualityCmaintenanceMCP server that enables AI agents to run a deterministic orchestration loop with decomposition, subagent execution, and review feedback across multiple LLM backends.61MIT
- AlicenseNot gradedqualityDmaintenanceEnables multi-model leader-worker agent orchestration, workflow execution, and deterministic validation via structured MCP tools.8 npmApache 2.0
- FlicenseNot gradedqualityAmaintenanceEnables multiple AI models to collaborate under a shared goal, architecture, plan, loops, sandbox, and acceptance criteria via a local-first MCP server.-