DACO — Declarative Agent & MCP Orchestration

One MCP endpoint to rule them all.

Instead of configuring N separate MCP servers in Claude Desktop, you configure one: DACO. It routes tool calls to the right backend, executes them in parallel when possible, and returns structured results.

The pattern is backend-agnostic. You define backends as simple modules (name prefix → handler function). DACO handles routing, parallel execution, error recovery, and the full MCP protocol over HTTP.

The LLM poses questions. DACO executes.

---

How It Works

Claude Desktop / Any MCP Client
        |
        | (single MCP connection)
        v
   +---------+
   |  DACO   |  Cloudflare Worker
   |  Router |  MCP Streamable HTTP
   +----+----+
        |
   +----+----+----+----+
   |    |    |    |    |
   v    v    v    v    v
 Backend A  B  C  D  ...

Each backend is a JS module exporting:

• TOOLS — array of MCP tool definitions
• callBackend(name, args, env) — handler function

DACO merges all tool lists, dispatches by name prefix, and adds meta-tools on top (daco_execute_parallel, daco_list_backends).

Current Backends (reference implementation)

This repo ships with 4 backends as a working example:

Backend	Prefix	What it does
Smart Rabbit	smart_rabbit_*	AI fitness program generation
PubMed	pubmed_*	Scientific literature search (NCBI)
Brave Search	brave_*	Real-time web search
FitLexicon	fitlexicon_*	Exercise database (873 exercises, 8 languages)

To adapt DACO to your own use case, replace these with your own backends. The orchestration layer doesn't care what the backends do.

Deploy

npm install
npx wrangler secret put BRAVE_API_KEY
npx wrangler secret put RAPIDAPI_KEY
npx wrangler deploy

Configure in Claude Desktop

{
  "mcpServers": {
    "daco": {
      "url": "https://your-worker.workers.dev/mcp",
      "transport": "http"
    }
  }
}

Or via local proxy:

{
  "mcpServers": {
    "daco": {
      "command": "npx",
      "args": ["mcp-remote", "https://your-worker.workers.dev/mcp"]
    }
  }
}

Adding a Backend

1. Create backends/my-service.js:

export const MY_SERVICE_TOOLS = [{
    name: 'myservice_do_thing',
    description: 'Does the thing',
    inputSchema: { type: 'object', properties: { query: { type: 'string' } }, required: ['query'] }
}];

export async function callMyService(name, args, env) {
    const res = await fetch('https://api.example.com/...', { ... });
    return JSON.stringify(await res.json());
}

2. Import in worker.js, add to ALL_TOOLS, add prefix routing in dispatchTool()

That's it. No config files, no plugin system. Just functions.

Parallel Execution

daco_execute_parallel fires multiple tool calls simultaneously:

daco_execute_parallel([
  { tool: "pubmed_search", arguments: { query: "hypertrophy" } },
  { tool: "brave_search", arguments: { query: "gym prices 2026" } },
  { tool: "myservice_do_thing", arguments: { query: "..." } }
])
→ All results returned in a single response

Related Projects

• immune — Adaptive memory system — learns patterns from every scan (+85% code quality)
• chimera — Bio-inspired 3-stage pipeline (Slime Mold → PRISM → Immune)
• spinal-loop — Neuromuscular-inspired agent routing (cheap models first)
• prism-framework — Multi-agent synthesis via native LLM stochasticity
• smartrabbit-mcp — AI workout generator MCP server (smartrabbitfitness.com)

---

MIT License — by Jacques Chauvin