Agent MCP: The Entire Worksona Workforce, Native to Claude Desktop

Agent MCP is a TypeScript MCP (Model Context Protocol) server that exposes the full Worksona agent library — over 100 agents across 19 professional categories — as native tools inside Claude Desktop.

MCP is an open protocol that lets external services present tools and resources to compatible AI clients. When you configure an MCP server in Claude Desktop, its capabilities appear alongside Claude's built-in tools. Agent MCP uses this to make every Worksona agent directly invokable from a conversation without any copy-paste, prompt engineering, or context switching.

What is it?

The server ships with three transport modes from a single TypeScript codebase:

Agent discovery is filesystem-driven. The build pipeline copies teams/ → agents/ via rsync, and the server uses glob to walk the directory tree at startup. No database required — the filesystem is the registry.

# Configure Claude Desktop (macOS)
# ~/.claude-desktop/config.json
{
  "mcpServers": {
    "worksona-agents": {
      "command": "/usr/local/bin/worksona-mcp-server"
    }
  }
}

After a restart, Claude Desktop loads the server and the full agent library is available as MCP tools.

Why is it useful?

Claude is a capable general-purpose assistant. But general-purpose means no specialization. When you need an AI that reasons like a cloud infrastructure engineer, or one that approaches a problem using Wardley Mapping, or one configured at doctoral depth with a deductive reasoning framework — the default model delivers inconsistent results unless you manually construct that context every time.

Agent MCP solves this by making specialization on-demand. The library's 19 categories cover the professional functions where structured expertise matters most: ai-machine-learning, board-of-directors, business-strategy, c-suite-executives, cloud-infrastructure, consultancy-agents, content-documentation, customer-engagement, data-engineering, devops-automation, and more.

The four core MCP tools the server exposes:

• activate_agent — engage a single named agent for a specific task
• coordinate_team — run a multi-agent workflow (sequential, parallel, review, or executive patterns)
• suggest_agents — receive AI-powered recommendations for which agents fit a given request
• list_agents — browse available agents by category

For complex work, coordinate_team is the key tool. It accepts a task and routes it through a coordinated agent workflow automatically:

Research the competitive landscape for AI coding assistants
→ coordinate_team activates: researcher, analyst, strategist, synthesizer
→ each agent processes in sequence and parallel per workflow definition
→ Claude assembles and presents the integrated output

How and where does it apply?

The Netlify Functions deployment mode is the infrastructure differentiator. Rather than requiring every user to install and maintain a local Node.js process, the MCP server can be hosted on Netlify and accessed via HTTP. This makes the agent library available to any HTTP-capable client — not just Claude Desktop.

The diagnostic tooling (agent-diagnostic-report.js) reflects the maturity of the library management approach. Running a health check against 100+ agent definitions catches malformed tool configurations, missing metadata, and invalid YAML before they surface as silent failures at runtime.

Agent MCP is the bridge between the Worksona agent ecosystem — built and maintained as a filesystem of structured definitions — and the Claude ecosystem, where those definitions become immediately usable capabilities. For teams that work in Claude Desktop daily, this turns a library of 100+ specialists into a toolbar.