A local switchboard for OpenAPI, GraphQL, and MCP

What it does

executor is a local runtime that ingests API definitions—OpenAPI, GraphQL, MCP, Google Discovery, or anything with a JSON schema—and exposes them as a unified, typed tool catalog. AI agents query it by intent, inspect schemas, and invoke functions through a TypeScript runtime or by treating executor itself as an MCP server. It runs locally with a web UI and CLI, auto-manages a background daemon, and handles authentication and execution pauses for manual approval.

The interesting bit

Instead of baking Stripe, GitHub, or Gmail adapters into every agent, executor centralizes them in one plugin-based catalog that any MCP-compatible client can consume. It turns your laptop into a private integration hub—complete with intent-based discovery and automatic port fallback—so Cursor, Claude Code, and friends all draw from the same well.

Key highlights

• First-class support for OpenAPI, GraphQL, MCP, and Google Discovery, with an open plugin system for anything JSON-schema-shaped.
• Serves itself as an MCP server, letting compatible agents share one catalog, auth config, and policy set.
• Typed TypeScript runtime with intent-based tool discovery and schema introspection.
• Local daemon with automatic port selection and a web UI for adding sources interactively.
• Executions can pause for auth or human approval and resume later via execution ID.

Caveats

• The repository description promises a “secure environment,” but the README never details the sandboxing or isolation model.
• “Policies” are mentioned as a shared asset, yet there is no explanation of how they are defined, enforced, or configured.

Verdict

Worth a look if you’re juggling multiple AI agents and tired of maintaining separate API adapters for each. Skip it if you need a managed, cloud-native integration platform—this is strictly localhost-first infrastructure.