← All posts

384 MCP Tools for Your AI Editor

I built 125 self-hosted developer tools. Each one runs as a single binary with an embedded SQLite database and a web dashboard. Feature flags, cron scheduling, webhook relay, alerting, expense tracking, secrets management, form builder, uptime monitoring, and a hundred more.

The problem was always the same: you install a tool, you open its dashboard, you click around. It works fine. But you're context-switching constantly. You're in Cursor writing code, then you tab over to check your feature flags, then back to Cursor, then over to the alert dashboard. It's death by a thousand tabs.

MCP changes this

The Model Context Protocol lets AI editors call external tools directly. Cursor, Claude Desktop, Windsurf, Cline. They all support it. You type a natural language request, the editor figures out which MCP tool to call, calls it, and shows you the result. No tab switching. No dashboard hopping.

So I built stockyard-mcp. One Go binary, 7MB, zero dependencies. It connects to your running Stockyard tools and exposes every single API endpoint as an MCP tool. 125 products, 384 MCP tools total.

What it looks like

You're in Cursor working on a feature. You want to know if the dark-mode feature flag is enabled. Instead of opening the SaltLick dashboard, you just type:

"Is the dark-mode feature flag enabled?"

Cursor calls the saltlick_list_flags MCP tool, which hits the SaltLick API running on port 9700, and shows you the result inline. Flag exists, enabled, 10% rollout.

Want to check your LLM spend? "What's my spend this month?" calls costcap_get_spend. Want to create a cron job? "Schedule a database backup every night at 2am" calls sundial_create_job. Want to check if your services are up? "Show me the health of all my monitored endpoints" calls outpost_list_monitors.

The adapter handles the translation. You don't need to know the API endpoints or the JSON schema. You just ask for what you want.

How it works

The architecture is simple. The adapter embeds a catalog of all 125 Stockyard products with their API schemas. When you start it, you tell it which tools are running and where:

stockyard-mcp --tools saltlick:9700,costcap:4100,sundial:9720

Or let it scan a port range:

stockyard-mcp --scan 4000-9900

It connects to each tool's /health endpoint to verify it's running, then serves MCP over stdio. Your editor talks JSON-RPC to the adapter, the adapter translates that into HTTP API calls to the right tool, and pipes the response back.

Setting it up

For Claude Desktop, add this to your config:

{
  "mcpServers": {
    "stockyard": {
      "command": "stockyard-mcp",
      "args": ["--tools", "saltlick:9700,costcap:4100"]
    }
  }
}

For Cursor, create .cursor/mcp.json:

{
  "mcpServers": {
    "stockyard": {
      "command": "stockyard-mcp",
      "args": ["--scan", "4000-9900"]
    }
  }
}

That's it. Restart the editor and every running Stockyard tool is available as a native MCP tool.

The bigger picture

This is the first step toward something I've been thinking about for a while. Right now the adapter exposes existing tool APIs. But the next step is an agent layer that can orchestrate actions across multiple tools from a single natural language prompt.

"Track my expenses, email me a weekly summary, flag anything over $500." That touches three different tools: Steward for expenses, Pony Express for email, Sentinel for alerts. Today you'd configure each one separately. Soon the agent will wire them together for you.

The tools are the platform. MCP, the agent, embeddable panels, custom dashboards. These are all just different interfaces to the same underlying tools. The more interfaces, the more useful the tools become.

If you're running any Stockyard tools, grab stockyard-mcp and try it. If you're not, browse the catalog and see if anything fits. The free tier gives you 5 items per tool, which is enough to try it out.

What tools would you want to talk to from your editor?