Neurosymbolic AI server combining Prolog's symbolic reasoning with MCP (github.com)

A new neurosymbolic AI server combines Trealla Prolog running in a WebAssembly/WASI runtime with the Model Context Protocol (MCP) to provide stateful, symbolic reasoning as a first-class tool for hybrid AI applications. The server exposes four core operations — loadProgram, runPrologQuery, saveSession and loadSession — and maintains a persistent Prolog session so knowledge bases and inference state survive between tool calls. All I/O is validated with Zod schemas to ensure type safety. The project is lightweight and developer-friendly (git clone, npm install, npm run build) and ships a sample MCP configuration that runs the server under Node with the four tools always allowed and a 15s timeout for agent integration (Cline/Roo/Copilot). For AI/ML practitioners this means easy, low-latency access to classical symbolic reasoning from model-driven agents: average query execution is ~12 ms (18 MB), program load ~8 ms (15 MB) and session save ~45 ms (22 MB). That performance plus persistent sessions makes it practical to maintain and evolve interpretable knowledge bases, run complex logical queries, and combine symbolic constraints with learned models in production agent pipelines. The Zod-based type safety and MCP integration lower integration risk, enabling reproducible, auditable reasoning steps inside LLM toolchains.