◆ MCP Local-first developer tool

Turn shell scripts into MCP tools your agents can call.

Conjure wraps any shell script in a typed, validated MCP tool — then runs locally so Claude Code, Cursor, Codex and friends can call it safely, with live output and full run history.

No cloud. No accounts. Runs on 127.0.0.1.
deploy-staging · output Success

Works with every popular MCP client

Claude CodeCursorCodexWindsurfClineVS CodeZedGemini CLI
The app

A calm, precise console for your tools.

Manage every tool, run it with live streaming output, and audit each call — all running locally.

Conjure Tools page — glass card grid of MCP toolsTools
How it works

From script to callable tool in a minute.

You already have the scripts. Conjure gives them types, guardrails, and a stdio MCP command.

1

Paste your script

Drop in any bash script. Conjure keeps it exactly as you wrote it.

#!/usr/bin/env bash
set -euo pipefail
echo "deploying $ENV"
kubectl apply -f ./$ENV
2

Define typed params

Add inputs with types, defaults, and validation. They compile to one JSON Schema — the single source of truth.

env   enum  · staging | prod
tag   string · default latest
wait  bool  · default true
3

Call it from your agent

Connect any MCP client and the tool shows up — runs stream live, every call is logged.

agent › run deploy-staging
        env=staging tag=v2.1
✓ rolled out · exit 0 · 3.8s
Use cases

What people conjure.

Pick one to see the exact flow — what the agent asks, the call it makes, and what the script streams back.

Connect

Install once, connect anywhere.

Run the UI with conjure, then point your MCP client at conjure --mcp.

Install and connect
Use one installed command for the UI and MCP stdio server.
conjure --mcp


        
Features

Built for running real commands, safely.

Everything you need to expose scripts to an agent without losing control.

Conjure your first tool.

Local, typed, and callable from your favorite agent in under a minute.