Crate embacle_server 

embacle-server

Unified OpenAI-compatible REST API + MCP server for embacle LLM runners.

Endpoints

• POST /v1/chat/completions — chat completion (streaming and non-streaming)
• GET /v1/models — list available providers and models
• GET /health — per-provider readiness check
• POST /mcp — MCP Streamable HTTP (JSON-RPC 2.0, via embacle-mcp)
• GET {UHP_BASE_PATH}/v1/uhp and friends — Unified Harness Protocol (default base /uhp)

Modules

• completions — chat completion handler with multiplex fan-out
• mcp_client — MCP stdio client pool for server-side tool execution (feature mcp-tools)
• models — model listing endpoint
• health — provider health checks
• auth — optional bearer token authentication
• streaming — SSE streaming for OpenAI-format responses
• provider_resolver — provider:model string routing
• router — Axum router wiring all endpoints (OpenAI + MCP)
• state — re-export of unified state from embacle-mcp
• runner — runner factory bridging to embacle core
• uhp — Unified Harness Protocol surface, mounted under its own base path

Modules

auth

Authentication for the OpenAI-compatible surface.

completions

health

mcp_client

MCP Client Pool

models

openai_types

provider_resolver

router

runner

state

streaming

uhp

Unified Harness Protocol (2026-08-11).