Skip to content
ToolHunt LogoToolHunt
Executor.sh - Open-Source MCP Gateway logo
Self-hostedAI InfrastructureAlt to Composio

Executor.sh - Open-Source MCP Gateway

Executor.sh is a fully open-source MCP gateway that connects AI agents to MCP servers, OpenAPI specs, GraphQL APIs, and custom tools through one endpoint you can self-host with Docker.

Interface preview

Executor.sh - Open-Source MCP Gateway user interface

Executor.sh is an open-source MCP gateway for wiring AI agents to external tools without a closed runtime or per-call lock-in. Configure integrations once—MCP servers, OpenAPI specs, GraphQL endpoints, or custom JS tools—and every MCP-compatible client (Claude Code, Cursor, Codex, and more) shares them through a single endpoint. Self-host with Docker, run the desktop app, or use the hosted cloud.

Key Features

  • One MCP Endpoint: Point Claude Code, Cursor, Codex, or any MCP client at Executor and reach every connected tool.
  • Protocol-Level Integrations: Import MCP servers, OpenAPI specs, GraphQL schemas, and custom tools; endpoints become typed tools automatically.
  • Context Efficiency: Agents see one gateway tool and load schemas only when needed, instead of dumping thousands of definitions into the prompt.
  • Semantic Discovery: Agents search tools by intent (for example, “send a Slack message”) rather than memorizing names.
  • SES Sandbox: Tool calls run in an isolated QuickJS sandbox; credentials are injected host-side and never enter the sandbox heap.
  • Policy Engine: Allow, require approval, or block tools based on semantics (GET vs DELETE, MCP destructiveHint, GraphQL mutations).
  • Flexible Deploy: Cloud, desktop (Mac/Windows/Linux), CLI (npm i -g executor), Docker self-host, or Cloudflare Workers.

Why Choose Executor.sh?

  • You want a self-hostable alternative to managed agent tool platforms like Composio.
  • You care about inspectable, MIT-licensed runtime code and zero per-call fees when self-hosted.
  • You already bring OpenAPI/GraphQL/MCP specs rather than only pre-built SaaS catalogs.
  • You need credential isolation and approval gates for destructive actions.

Docker Deployment

Quick start with Docker:

docker run -d \
  --name executor-selfhost \
  -p 4788:4788 \
  -v executor-data:/data \
  ghcr.io/rhyssullivan/executor-selfhost:latest

Or with Docker Compose:

services:
  executor:
    image: ghcr.io/rhyssullivan/executor-selfhost:latest
    container_name: executor-selfhost
    restart: unless-stopped
    ports:
      - "4788:4788"
    volumes:
      - executor-data:/data
    environment:
      - PORT=4788
      - EXECUTOR_DATA_DIR=/data

volumes:
  executor-data:

Open http://localhost:4788. The first account becomes the owner. Connect agents with:

npx add-mcp https://executor.yourdomain.com/mcp

Getting Started

  1. Deploy with Docker or try the hosted cloud.
  2. Add integrations (MCP servers, OpenAPI, GraphQL).
  3. Point your MCP client at the Executor endpoint.
  4. Set policies for sensitive or destructive tools.
  5. Run multi-agent workflows with one shared gateway.

Full Setup Guide

For a full comparison with Composio and a step-by-step Docker walkthrough (including Dokploy), read: Executor.sh: Self-Hosted Open-Source Composio Alternative.

Executor.sh - Open-Source MCP Gateway logo
Self-hostedAI Infrastructure

Executor.sh - Open-Source MCP Gateway

Executor.sh is a fully open-source MCP gateway that connects AI agents to MCP servers, OpenAPI specs, GraphQL APIs, and custom tools through one endpoint you can self-host with Docker.

Alternative to Composio

Key features

  • MCP gateway for any compatible agent
  • OpenAPI, GraphQL, and MCP integrations
  • Context-efficient dynamic tool loading
  • SES sandbox with host-side credentials
  • Policy engine for safe vs destructive calls
  • Docker, CLI, desktop, and Cloudflare deploy

More AI Infrastructure tools

Hindsight - Agent Memory System for AI logo
Self-hostedAI Infrastructure

Hindsight - Agent Memory System for AI

Hindsight is an open-source agent memory platform that helps AI agents retain, recall, and reflect so they learn over time instead of starting every session from zero.

Alternative to LangChain

View detailsVisit site
Riwaq - Multi-Tenant AI Agent Backend logo
Self-hostedAI Infrastructure

Riwaq - Multi-Tenant AI Agent Backend

Riwaq is a Docker-first multi-tenant agent platform with RAG, private memory, self-learning from feedback, reminders, and OpenAI-compatible APIs.

Alternative to LangChain

View detailsVisit site