Anubis MCP

Anubis MCP: Bridging Large Language Models and External Systems

Anubis MCP is a purpose‑built Model Context Protocol implementation written in Elixir. It resolves the friction that developers face when they need to expose native services—such as data stores, micro‑services, or custom business logic—to AI assistants. By implementing both client and server sides of the MCP specification, Anubis allows a single codebase to act as a bridge: an LLM can call tools, retrieve prompts, and sample from external models while the server guarantees fault tolerance, concurrency, and a clean API surface.

The server’s core value lies in its lightweight yet powerful tool registration system. A developer can declare tools declaratively, specifying input schemas, descriptions, and annotations. Once registered, the LLM can invoke these tools through standard MCP calls; the server then executes Elixir callbacks, logs usage, and returns results in a predictable JSON format. This pattern eliminates the need for bespoke HTTP endpoints or custom SDKs, letting developers focus on business logic rather than protocol plumbing.

Key capabilities include:

• Tool Management: Register, update, and expose arbitrary tools with rich input validation.
• Session State: Persist per‑client state via assigns, enabling counters, caches, or context tracking across calls.
• Transport Flexibility: Support for streamable HTTP and Server‑Sent Events, making the server usable in both web and embedded environments.
• Client SDK: A matching client module that handles protocol negotiation, request routing, and response parsing, so callers can simply invoke without worrying about low‑level details.

Typical use cases span from simple echo services to complex workflow orchestration. For example, a finance application can expose an “evaluate risk” tool that pulls market data and returns a score; an e‑commerce platform can provide “recommend products” as a tool that queries its catalog. In each scenario, the LLM can seamlessly orchestrate these tools as part of a conversational chain, while the server ensures reliability and scalability through Elixir’s BEAM runtime.

Anubis stands out by coupling the MCP spec with Elixir’s concurrency model. Each tool invocation runs in its own lightweight process, allowing thousands of parallel calls without blocking. Moreover, the library’s registry and supervision strategy mean that crashes are isolated; a misbehaving tool does not bring down the entire server. For developers already invested in Elixir, this means minimal friction and maximum confidence that their MCP integration will scale with production traffic.