Stop duct-taping docs, SDKs, and MCP servers together

What it does

Cortex ingests API specifications—OpenAPI, AsyncAPI, GraphQL, gRPC, and OpenRPC—alongside custom Markdown. From a single cortex.config.yml, it generates interactive static documentation, typed client SDKs in eleven languages, and a Model Context Protocol (MCP) server. The goal is to stop coordinating separate toolchains every time an API changes.

The interesting bit

The MCP server output is the unusual kicker: it packages your specs, SDK guides, and project docs into typed tools that AI agents can call directly. Cortex also merges multiple protocols into a single per-language SDK if the package name matches, which is either convenient or terrifying depending on your API surface area.

Key highlights

• Generates SDKs for TypeScript, Python, Go, Java, Kotlin, Ruby, PHP, C#, Rust, C++, and C from one config.
• Produces static HTML documentation with interactive API references; the deployed site needs no runtime server.
• Creates an MCP server embedding callable tools, raw specifications, and Markdown guides for AI agents.
• Supports sparse Eta template overrides instead of maintaining full custom generator forks.
• Merges heterogeneous sources—REST, WebSocket, GraphQL, gRPC, JSON-RPC—into unified language packages.

Caveats

• The generated MCP server exposes gRPC Protocol Buffer files as read-only resources; it does not generate callable gRPC tools for agents.
• custom_head_html is injected raw without sanitization, so trusting your input is mandatory.
• Generated SDKs still rely on each target language’s native compiler and package manager.

Verdict

Teams managing multi-protocol APIs who are tired of syncing docs, SDKs, and AI context manually should look here. If you only need a static docs site or a single-language SDK, it is probably overkill.