Jcode, developed by 1jehuang, is an MCP server tool primarily built using Rust. It functions as an AI agent skill for coding assistants and a coding-agent harness, connecting to a central server process that manages sessions and state. TUI clients communicate over a Unix socket, while it integrates with Model Context Protocol servers and custom OpenAI-compatible endpoints. Jcode supports over 30 Large Language Model providers, offering a versatile platform for AI-assisted development workflows.

What does Jcode connect to?

Jcode connects to a central server process managing sessions and state, with TUI clients communicating over a Unix socket. It integrates with Model Context Protocol (MCP) servers and custom OpenAI-compatible endpoints. The system supports over 30 Large Language Model providers, including NVIDIA NIM DeepSeek-V4 models. For LLM providers, Jcode uses HTTP/HTTPS, accepting plain HTTP for localhost and private LAN IPs, but rejecting public remote HTTP connections. Compatible AI clients include Claude Code, Codex CLI, ChatGPT, OpenAI, Gemini, and Antigravity.

Core features and tools

Jcode provides several callable tools for agent workflows. These include explicit memory tools for active information search or storage, and session search for Retrieval-Augmented Generation on previous interactions. Display modes are managed via the /alignment command, while server hot-reloading is handled by the /reload command. Authentication is initiated with jcode login, and a detached daemon can be spawned using jcode serve. An ambient agent also gardens the memory graph and executes scheduled queue items.

Who is it for?

Jcode targets developers seeking greater control over agent workflows and those needing high performance and resource efficiency for multi-session coding. It is also suitable for tool builders testing agent harnesses, as well as beginners or developers working in minimal setups. Its capabilities extend to enabling multiple AI agents to collaborate on the same codebase and managing semantic memory. This is not the right fit if you require direct public remote HTTP connections to LLM providers.

How to get started

Access is granted through a standard OAuth authorization flow or via API keys. API keys can be set using environment variables like ANTHROPIC_API_KEY or OPENAI_COMPAT_API_KEY.

Use cases

• Multi-session agent workflows and swarm collaboration: enabling multiple AI agents to work concurrently on the same codebase, with the server detecting and notifying agents of file edit conflicts.
• Semantic memory and RAG: utilizing a semantic memory system that embeds conversation turns as vectors, allowing agents to actively search or store information and perform RAG on previous sessions.
• Self-development/Self-editing: allowing the AI to edit, build, test, and reload its own source code, facilitating iterative development.

Jcode by 1jehuang is a Rust-based MCP server tool designed to enhance AI coding assistant capabilities. It offers advanced features like multi-session agent workflows, semantic memory, and AI self-development, supporting numerous LLM providers via HTTP/HTTPS and Unix sockets. Its authentication system uses OAuth or API keys, configured through environment variables or specific commands. Jcode caters to developers seeking fine-grained control and performance in AI agent interactions.