Overview

Relevant source files

• .claude/skills/emulate/SKILL.md
• .gitignore
• .vscode/settings.json
• AGENTS.md
• CLAUDE.md
• RUNNING.md
• apps/cli/package.json
• apps/cloud/package.json
• apps/cloud/src/web/pages/login.tsx
• apps/desktop/package.json
• apps/docs/.gitignore
• apps/docs/README.md
• apps/docs/concepts/connections.mdx
• apps/docs/concepts/integrations.mdx
• apps/docs/concepts/policies.mdx
• apps/docs/docs.json
• apps/docs/favicon.png
• apps/docs/hosted/cloud.mdx
• apps/docs/hosted/cloudflare.mdx
• apps/docs/hosted/docker.mdx
• apps/docs/images/desktop-app.png
• apps/docs/index.mdx
• apps/docs/local/cli.mdx
• apps/docs/local/desktop.mdx
• apps/docs/mcp-proxy.mdx
• apps/docs/style.css
• apps/local/package.json
• apps/marketing/astro.config.mjs
• apps/marketing/package.json
• apps/marketing/public/og-image.png
• apps/marketing/public/pattern-graph-paper.svg
• apps/marketing/src/components/LegalLayout.astro
• apps/marketing/src/components/animated-beam-demo.tsx
• apps/marketing/src/components/context-bloat-demo.tsx
• apps/marketing/src/components/ui/animated-beam.tsx
• apps/marketing/src/env.d.ts
• apps/marketing/src/layouts/Layout.astro
• apps/marketing/src/lib/github.ts
• apps/marketing/src/lib/utils.ts
• apps/marketing/src/middleware.ts
• apps/marketing/src/pages/index.astro
• apps/marketing/src/styles/global.css
• apps/marketing/wrangler.toml
• bun.lock
• e2e/CHANGELOG.md
• e2e/package.json
• e2e/scripts/serve.ts
• package.json
• packages/plugins/desktop-settings/package.json
• packages/react/package.json
• scripts/bootstrap.ts
• scripts/migrate-schema-class.ts
• tsconfig.json

Executor is an open-source integration layer for AI agents. It provides a unified platform to configure and manage various integration types—including MCP servers, OpenAPI specs, and GraphQL APIs—and exposes them to agents through a single Model Context Protocol (MCP) interface apps/marketing/src/pages/index.astro23-26

The project is structured as a TypeScript monorepo using Bun and Turbo, designed to run across diverse environments: local machines (CLI/Desktop), self-hosted infrastructure (Docker/Cloudflare), and a managed Cloud offering package.json23-30 apps/marketing/src/pages/index.astro27-37

System Architecture Overview

Executor bridges the gap between heterogeneous API protocols and AI agent interfaces. It employs a layered architecture where a core SDK manages the lifecycle of integrations, while specific host applications package this logic for different runtime environments.

From Natural Language to Code Entities

Sources: package.json5-12 apps/marketing/src/pages/index.astro42-46 AGENTS.md67-78

---

Monorepo Organization

The codebase is partitioned into four primary functional areas to ensure portability and separation of concerns AGENTS.md67-78:

Category	Path	Description
Apps	apps/*	Product entry points: cli, cloud, desktop, and local bun.lock31-185
Core	packages/core/*	Fundamental logic: sdk, fumadb (storage), and execution engine AGENTS.md68-72
Plugins	packages/plugins/*	Protocol drivers (OpenAPI, MCP, GraphQL) and credential providers AGENTS.md73-74
Kernel	packages/kernel/*	Sandboxed runtimes for executing tool code safely AGENTS.md77

Monorepo Component Mapping

Sources: bun.lock31-185 package.json23-30 AGENTS.md67-78

---

Core Concepts

Executor operates on a set of domain-specific entities that define how tools are discovered and executed:

• Integration: A definition of a service (e.g., a specific OpenAPI spec or MCP server path) apps/marketing/src/pages/index.astro43
• Connection: A configured instance of an integration, often carrying specific authentication credentials apps/marketing/src/pages/index.astro44
• Policy: A set of rules determining if a tool execution is allowed, blocked, or requires manual approval apps/marketing/src/pages/index.astro45
• Host: The runtime environment (CLI, Desktop, Cloud) that provides the Executor instance and exposes the API AGENTS.md78

For a full definition of these terms and their code identifiers, see Key Concepts and Terminology.

---

Development and Deployment

The project uses Effect for functional programming patterns and Vitest for testing AGENTS.md10 package.json96 The release process is managed via Changesets, producing NPM packages, CLI binaries, and Electron installers package.json54-66

• Getting Started: For environment setup and booting the dev servers, see Getting Started.
• Architecture: For a deep dive into the internal layers, see Core Architecture.
• Plugins: To understand how to add new protocols or secret managers, see Plugins and Integrations.
• Hosts: For details on specific application targets like the Electron app or Cloudflare Workers, see Host Applications.

Sources: package.json32-69 AGENTS.md1-15 bun.lock1-30