conjure — Build Claude Code Extensions from Plain Language
Build Claude Code skills, agents, hooks, and MCP servers from plain-language descriptions. Describe what you want — conjure writes production-ready artifacts instantly. No boilerplate. No manual wiring.
claude plugin marketplace add john-farina/claude-plugins
claude plugin install conjure@john-farina
→ conjure.dev landing page
What is conjure?
Conjure is a Claude Code plugin that builds the four types of Claude Code extensions:
- Skills — slash commands you type (
/my-command)
- Agents — autonomous workers Claude spawns programmatically
- Hooks — event-driven shell scripts that fire automatically
- MCP Servers — live connections to external APIs, databases, and tools
Describe what you want in plain language. Conjure classifies the extension type, asks targeted questions, checks whether a published solution already exists, and writes a production-ready artifact. Every command also handles updates — no need to say "update."
How to build a Claude Code skill
/conjure-skill "summarize my git changes and write a commit message in my team's style"
Conjure writes the .md command file with proper frontmatter, tool declarations, argument handling, and phased instructions. Immediately usable in the same session.
How to build an MCP server for Claude Code
/conjure-mcp "connect Claude to our internal Postgres database"
Conjure searches smithery.ai and mcp.so first. If no match exists, it scaffolds a full server in Node.js, Python, or Go with typed tools and registers it in settings.json.
How to create a Claude Code agent
/conjure-agent "review PRs and post feedback in my voice"
Conjure handles architecture decisions (orchestrator vs. worker), tool selection, maxTurns, permissionMode, and model config. Writes the definition to ~/.claude/agents/.
How to add a Claude Code hook
/conjure-hook "run prettier on every file Claude edits"
Conjure writes the shell script and the settings.json entry. Supports all hook events: PreToolUse, PostToolUse, SessionStart, UserPromptSubmit, Stop.
Commands
| Command | What it does |
|---|
/conjure | Auto-detects type from description and delegates |
/conjure-skill | Create or update a slash command / workflow |
/conjure-agent | Create or update an autonomous worker |
/conjure-hook | Create or update an event-driven hook |
/conjure-mcp | Create or update an MCP server |
/conjure-config | Set preferences all conjure commands follow automatically |
/conjure-help | Full reference and decision guide |
conjure-config — preferences that apply everywhere
Run /conjure-config once. Tell conjure your output paths, VCS tool, author name, naming conventions, required frontmatter — anything. It writes plain-language instructions to ~/.claude/conjure-config.md (global) and .claude/conjure-config.md (project-local). Every conjure command reads these automatically and skips questions you've already answered.
License
MIT
