SuperAgenticMCP

Patch every agent into every tool.

When to use this skill

• Orchestrate multiple agents (planner / workers / critic) against a set of MCP servers
• Route tool calls by capability with latency as a tie-breaker
• Run a local switchboard that racks/unracks servers and regenerates config
• Observe live agent→server traffic (patch board + JSONL log + run receipt)
• Persist run artifacts into a linked memory-scope constellation
• Generate or consume a superagentic.json router config

How to run the MCP server

pip install superagenticmcp
# or from source:
pip install -e .

superagenticmcp

MCP client config:

{
  "mcpServers": {
    "superagenticmcp": {
      "command": "superagenticmcp",
      "args": []
    }
  }
}

Board mode (when implemented):

superagenticmcp up --board
# board on http://localhost:7420

Core ideas

1. Rack — toggle MCP servers online; they appear on the board and become routable
2. Plan — planner decomposes the task order into steps across servers
3. Route — capability match first, latency second, fallback when a server is offline
4. Observe — every call is a JSONL line; the same stream feeds the log and the board
5. Remember — completed runs file artifacts into a linked memory graph

Principles

1. Observable by default — routing and failures must be visible (log + board + receipt)
2. Config as truth — rack state and superagentic.json stay in sync
3. Fail closed on budgets — a run that hits its call budget stops and says so
4. Demo is not runtime — the HTML hero is for design feel; production path is the Python MCP server
5. Additive surface — keep the public MCP tool surface stable and clear

Interactive shell

No install required:

https://htmlpreview.github.io/?https://github.com/ANAMIZED/SuperAgenticMCP/blob/main/superagenticmcp.html