Building an MCP Server for Pickaxe: Managing AI Agents Through Natural Language

If you’re building AI agents on Pickaxe, you know the drill: log into the dashboard, click through menus, copy-paste content, check analytics, manage users. Repeat daily.

What if you could do all of that through conversation instead?

That’s exactly what I built with mcp-pickaxe—an MCP server that connects Claude directly to the Pickaxe platform. No more context-switching between your AI assistant and the Pickaxe dashboard.

What is MCP?

https://www.youtube.com/watch?v=aZLr962R6Ag

https://www.youtube.com/watch?v=MZ-u8r\_Mbsg

Model Context Protocol (MCP) is Anthropic’s standard for connecting AI assistants to external tools and data sources. Think of it as a universal adapter that lets Claude interact with your systems directly.

What Can You Do With mcp-pickaxe?

Once installed, you can manage your entire Pickaxe operation through conversation:

Analyze Agent Performance

• “Show me the last 20 conversations from my support agent”

• “What questions are users asking that my agent can’t answer?”

This is where it gets powerful. Instead of manually reviewing chat logs, you can ask Claude to identify patterns, find knowledge gaps, and suggest improvements to your agent’s knowledge base.

Manage Knowledge Bases

• “Create a new document called ‘FAQ’ with this content…”

• “Connect the FAQ document to my customer support agent”

• “List all documents in my knowledge base”

No more navigating to the documents section, clicking create, pasting content, then separately linking it to agents. One command handles the entire workflow.

Handle User Management

“Show me all users and their usage stats”

“Create a new user with email [email protected] and give them access to the Pro product”

“Send invitations to these emails: [list]”

Bulk operations that would take dozens of clicks become single requests.

Multi-Studio Operations

If you manage multiple Pickaxe studios (production, staging, development), you can work across all of them:

“List all users in my staging studio”

“Compare the documents between production and staging”

The Technical Details

The server provides 17 tools across six categories:

Category	What You Can Do
Studios	List configured studios, switch between them
Chat History	Fetch and analyze agent conversation logs
Documents	Create, list, get, delete, connect/disconnect to agents
Users	Create, list, get, update, delete, invite
Products	List available products and bundles
Memory	List memory schemas, retrieve user memories

Installation

The server requires Node.js 18+ and your Pickaxe Studio API key.

git clone https://github.com/aplaceforallmystuff/mcp-pickaxe.git
cd mcp-pickaxe
npm install
npm run build

Then configure your MCP client (Claude Desktop or Claude Code) with your API key.

Full installation instructions are in the repository.

Why I Built This

I use Pickaxe to build AI agents for various projects. The platform is excellent, but I found myself constantly switching between Claude (where I do most of my thinking and writing) and the Pickaxe dashboard (where I manage agents).

MCP eliminates that friction. Now my AI workflow and my agent management live in the same conversation.

The same principle applies to everything I’m building lately: integration beats capability. The best tool isn’t necessarily the most powerful one—it’s the one that connects seamlessly to your existing workflow.

Get Started

The code is open source under MIT license:

If you’re using Pickaxe and Claude, give it a try. And if you build something interesting with it, I’d love to hear about it.

Resources & Further Reading

• Model Context Protocol Documentation: https://modelcontextprotocol.io

• Anthropic Claude: https://claude.ai

• Pickaxe Platform: https://pickaxe.co

• mcp-pickaxe GitHub Repository: https://github.com/aplaceforallmystuff/mcp-pickaxe