Skills — Slash Commands
A /command you type to trigger a workflow. Claude Code skills handle multi-phase tasks, argument parsing, tool use, and output formatting.
Claude Code Plugin
Build Claude Code extensions
from plain language
Describe what you want — a skill, agent, hook, or MCP server. Conjure classifies the type, asks what it needs, and writes production-ready code. No boilerplate. No manual wiring.
Install
claude plugin marketplace add john-farina/claude-plugins
claude plugin install conjure@john-farina
Free & open source · Works with any project · View on GitHub
Build any Claude Code extension from a description
Four extension types. One plain-language command. Conjure picks the right one and builds it.
Agents — Autonomous Workers
Isolated workers Claude spawns programmatically. Agents run with their own context, tool restrictions, model selection, and turn limits.
/conjure-agentHooks — Event Triggers
Shell scripts that fire automatically on lifecycle events — before tool calls, after edits, on session start, on every prompt. Blocking or non-blocking.
/conjure-hookMCP Servers — Live Connections
Connect Claude to any external system — databases, APIs, internal tools. Scaffolds a full server with typed tools, resources, and settings.json registration.
/conjure-mcpHow conjure works
Start with /conjure — it handles the rest.
1
Describe what you want
Type /conjure followed by a plain-language description — or just the name of an existing extension to update it. Conjure infers the extension type automatically.
2
Conjure checks for existing solutions
For MCP servers and agents, conjure searches smithery.ai, mcp.so, and GitHub before building. If a strong match exists, install it directly instead of starting from scratch.
3
Targeted intake questions
Conjure asks only what it needs — tool access, trigger conditions, output format, storage location. Your preferences from /conjure-config are applied automatically so you skip questions you've already answered globally.
4
Production-ready artifact, installed immediately
The extension is written, validated, and ready to use before conjure finishes. Skills are invocable in the same session. Hooks are wired into settings.json. MCP servers are registered and running.
Commands
Seven commands — start with /conjure if you're unsure which to use.
| Command | What it builds |
|---|---|
| /conjure | Auto-detects type from your description and delegates to the right command |
| /conjure-skill | Claude Code slash commands and multi-phase workflows |
| /conjure-agent | Autonomous worker definitions with tool restrictions and model config |
| /conjure-hook | Event-driven shell scripts wired into settings.json |
| /conjure-mcp | Model Context Protocol servers in Node.js, Python, or Go |
| /conjure-config | Set plain-language preferences all conjure commands follow automatically |
| /conjure-help | Full reference — all argument syntax, hook events, update mode, examples |
Frequently asked questions
Common questions about building Claude Code extensions with conjure.
How do I install conjure?
Run two commands in your terminal:
Start a new Claude Code session. Type
claude plugin marketplace add john-farina/claude-pluginsclaude plugin install conjure@john-farinaStart a new Claude Code session. Type
/conjure to begin.
How do I create a Claude Code skill (slash command)?
Run
/conjure-skill and describe the command you want — what triggers it, what it should do, and any tools it needs. Conjure writes the .md command file with proper frontmatter, tool declarations, and phased instructions. It's immediately usable in the same session.
How do I build an MCP server for Claude Code?
Run
/conjure-mcp and describe the external system you want to connect — a database, a REST API, an internal tool. Conjure searches smithery.ai and mcp.so first to see if a matching server already exists. If not, it scaffolds a working server with typed tools and registers it in settings.json.
How do I add a hook to Claude Code?
Run
/conjure-hook and describe the trigger and behavior you want. For example: "run prettier on every file Claude edits" or "inject my coding standards on every prompt." Conjure writes the shell script and adds the hook entry to settings.json automatically.
What is the difference between a skill and an agent?
A skill is a slash command a user types — it runs in the main Claude context. An agent is an autonomous worker Claude spawns programmatically, with its own isolated context, separate tool permissions, and turn limits. Use skills for user-triggered workflows; use agents for parallel or delegated subtasks.
Can conjure update extensions I already have?
Yes. Every conjure command auto-detects create vs. update mode. Provide the name of an existing extension and conjure reads it in full before refining it. You never need to say "update" — conjure figures it out from the name.
Does conjure work with any project or just specific languages?
Conjure works with any project. The extensions it builds are Claude Code artifacts —
.md files, shell scripts, JSON config, and MCP servers. They're language-agnostic and work regardless of what codebase Claude Code is running against.
How does /conjure-config work?
Run
/conjure-config for interactive setup. Conjure asks about your preferences — output paths, naming conventions, required frontmatter, VCS tool, author name — and writes plain-language instructions to ~/.claude/conjure-config.md (global) or .claude/conjure-config.md (project-local). Every conjure command reads these automatically. Project-local wins on conflicts.
Start building in 30 seconds
Two commands. Then describe what you want.
Install conjure
claude plugin marketplace add john-farina/claude-plugins
claude plugin install conjure@john-farina