# AI-Powered Pickaxe Development Pipeline with Claude Code + MCP

**URL:** <https://community.pickaxe.co/t/ai-powered-pickaxe-development-pipeline-with-claude-code-mcp/7738>\
**Category:** General\
**Created:** [December 9, 2025, 12:32pm UTC](https://community.pickaxe.co/t/ai-powered-pickaxe-development-pipeline-with-claude-code-mcp/7738 "2025-12-09T12:32:42Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Jim](https://yyz2.discourse-cdn.com/flex004/user_avatar/community.pickaxe.co/jim/32/2820_2.png) [@Jim](https://community.pickaxe.co/u/Jim)\
**Post date:** [December 9, 2025, 12:32pm UTC](https://community.pickaxe.co/t/ai-powered-pickaxe-development-pipeline-with-claude-code-mcp/7738/1 "2025-12-09T12:32:42Z")

</div>

I wrote a couple weeks ago about my Pickaxe MCP server for Claude Code. Since then I’ve been working more on automating the analysis and creation process of Pickaxes:

Creating and maintaining agents manually through the dashboard works fine for one-offs, but when you’re iterating on prompts, managing knowledge bases, and auditing performance across dozens of agents, it gets tedious fast.

So I built a system that lets me manage Pickaxe from my terminal using Claude Code and the Model Context Protocol (MCP).

**What I Built**

1. MCP Server (mcp-pickaxe) - Connects Claude directly to the Pickaxe API
2. Specialized Agents - Claude subagents that handle specific tasks
3. Automation Skills - Reusable workflows for common operations

**The MCP Server**

Published on npm as mcp-pickaxe. It gives Claude access to:

- Chat history analysis
- Document/KB management (create, connect, disconnect)
- User management and usage stats
- Multi-studio support (switch between studios in one session)

What it CAN’T do (yet): Create pickaxes themselves - that still requires the dashboard.

**The Agent Pipeline**

I have 4 specialized Claude agents that work together:

| Agent | Purpose |
| --- | --- |
| Pickaxe Auditor | Reviews studio health - activation rates, KB quality, pricing gaps |
| Pickaxe Planner | Designs new agents with implementation-ready specs |
| Pickaxe Builder | Walks through dashboard configuration section by section |
| Pickaxe Uploader | Handles document creation and KB connections |

Example Workflow: Gap Analysis → New Agent

1. “Audit my RRHUB studio” → Auditor analyzes users, chat history, KB quality
2. Auditor identifies: “Users asking about document analysis but no agent handles it”
3. “Plan a Document Analyzer agent” → Planner creates full spec
4. Spec includes: system prompt, token settings, KB connections, test cases
5. “Upload these KB documents” → Uploader pushes to Pickaxe (75% success rate, there’s still some manual connection and oversight needed)
6. I manually create the agent in dashboard using the spec (copy-paste)

**The Latest Addition: Comet Skill Creator**

I just added a Claude skill that generates shortcuts for Perplexity’s Comet browser. Now I can say “create a Comet shortcut that automates Pickaxe agent creation” and get a ready-to-paste automation prompt. Now I invoke the skill in Comet, attach the spec file and away it goes.

The dream: eventually automate the dashboard itself via browser automation (or maybe more facility to do this in the API?). Here’s a video of the pickaxe creation being automated in Comet while I watch (albeit a bit slow)

[](https://www.youtube.com/watch?v=UFugdv3hHic)

---

<div class="post-metadata">

**Author:** ![kenlyle](https://avatars.discourse-cdn.com/v4/letter/k/ac8455/32.png) [@kenlyle](https://community.pickaxe.co/u/kenlyle)\
**Post date:** [February 11, 2026, 4:31pm UTC](https://community.pickaxe.co/t/ai-powered-pickaxe-development-pipeline-with-claude-code-mcp/7738/2 "2026-02-11T16:31:03Z")

</div>

Holy Cow, @jim ! Incredible. I am planning a social media generator where each vertical might have its own KB, including glossary/terms.

I’d love to play with the other tooling, if there are parts that do not rely on the API.