Barbacane is a spec-driven API gateway developed by Barbacane Dev, primarily built with Rust. Its core compiler and runtime are in Rust, making up 88.2% of the codebase. This free program allows plugins to be developed in any language compiling to WebAssembly, including Rust, Go, or AssemblyScript. Barbacane compiles OpenAPI or AsyncAPI specifications into a sealed .bca artifact, which is then served by its data plane. It functions as both an MCP server and an AI gateway, connecting various backend services.

What does Barbacane connect to?

Barbacane connects to diverse backend services, including HTTP, Lambda, S3, Kafka, NATS, LLMs, and WebSocket backends. It operates as a spec-driven API gateway, categorizing itself as an MCP server and an AI gateway. This allows it to centralize access and management for a variety of service types, routing requests based on defined specifications. Its architecture supports a broad ecosystem of modern and traditional services.

Core features and tools

Barbacane exposes API operations as AI-callable tools via a JSON-RPC 2.0 MCP endpoint at POST /__barbacane/mcp. It supports specific callable JSON-RPC methods: initialize, notifications/initialized, tools/list, tools/call, and ping. Each operation in the OpenAPI spec becomes a typed tool, named from its operationId and described by its summary. Compatible AI clients include OpenAI, Anthropic, and Ollama, unified behind an OpenAI-compatible surface. Inbound MCP clients like Claude Code are confirmed compatible.

Who is it for?

Barbacane is for developers and teams aiming to expose existing APIs to AI agents as typed MCP tools or route outbound LLM calls. It helps in making APIs discoverable and callable by AI agents, and in managing requests to various LLM providers. This program is particularly useful for those needing to integrate AI capabilities with their existing service infrastructure. Skip this if you do not need to expose APIs to AI agents or route LLM requests.

How to get started

Access is granted through a standard OAuth authorization flow. HTTP serves as the transport protocol for the MCP endpoint.

Use cases

• Exposing APIs to AI Agents: Developers define their API using OpenAPI/AsyncAPI, enable MCP, and Barbacane automatically makes each operation available as a typed tool for AI agents to discover and call.
• Routing Outbound LLM Requests: Applications send OpenAI-format requests to Barbacane, which then intelligently routes them to various LLM providers (e.g., OpenAI, Anthropic, Ollama) with features like provider fallback, policy routing, and prompt guarding.
• Spec-Driven API Gateway Deployment: Teams can define their API with Barbacane extensions, compile it into a .bca artifact, and deploy it to achieve routing, validation, and authentication with compile-time safety and extensibility via WebAssembly plugins.

Barbacane offers a Rust-based, spec-driven API gateway for integrating AI agents with existing services and managing LLM traffic. Its WebAssembly plugin support and multiple authentication methods provide adaptability for various deployment scenarios. The program unifies access to different LLM providers, presenting a consistent interface for AI clients.