forge-orchestrator
Forge Orchestrator: Multi-AI task coordination. File locking, knowledge capture, drift detection. Rust.
claude mcp add --transport stdio nxtg-ai-forge-orchestrator ./target/release/forge-orchestrator \ --env LOG_LEVEL="info" \ --env FORGE_STATE_DIR="Default state directory (e.g., .forge/)."
How to use
forge-orchestrator is a Rust-based MCP server that coordinates multiple AI tools and adapters (Claude Code, Codex CLI, Gemini CLI) within a single, governed workflow. It provides file locking, knowledge capture, task planning, drift detection, and multi-tool orchestration all through a unified state directory (.forge). Tools connect to this MCP server via standard MCP IO channels, share a single source of truth, and operate without conflicting edits. You’ll use it to manage cross-tool tasks, monitor lock states, and track decisions across sessions. The system exposes a plan-driven workflow where tasks are decomposed, dependencies tracked, and tools assigned automatically or by human-in-the-loop, with drift detection to flag divergences from specs.
To use it, start the forge-orchestrator server in the environment where you’ve installed it. Tools connect to the MCP server to read and write their configured state and artifacts. You can inspect the live state via the provided dashboard and interact with the plan using the forge CLI (init, plan --generate, run, dashboard, status). This MCP server serves as the policy core that enforces ordering, locking, and knowledge consolidation while adapters present state through their respective interfaces (L1 safety plugin, L3 UI, and MCP stdio).
How to install
Prerequisites:
- Rust toolchain (stable) installed on your machine
- curl and Git
- A Unix-like shell environment (Linux/macOS) or WSL on Windows
Install and run from source:
-
Clone the repository: git clone https://github.com/nxtg-ai/forge-orchestrator cd forge-orchestrator
-
Build the release binary: cargo build --release
-
Run the MCP server locally: ./target/release/forge-orchestrator
The server will start and listen for MCP tool connections on the default ports
Optional one-liner install script (per project README): curl -fsSL https://forge.nxtg.ai/install.sh | sh forge init
Notes:
- The orchestrator is a single binary (~4MB) with no runtime dependencies beyond the system Rust runtime.
- After starting, you can initialize a project directory with forge init and begin planning tasks with forge plan --generate.
Additional notes
Tips and common considerations:
- State is stored in .forge/; ensure this directory is writable and backed up if needed.
- The MCP server enforces file locking to prevent concurrent edits across tools.
- Use forge dashboard for a live view of task status, locks, and tool activity.
- If you integrate with external UI (Forge UI) or plugins, ensure their adapters are compatible with the MCP stdio and filesystem conventions.
- Environment variables can tune behavior (e.g., FORGE_STATE_DIR to override the default state directory, LOG_LEVEL for verbosity).
- When upgrading, debris from previous sessions may linger in .forge/knowledge; consider pruning or archiving stale entries if needed.
Related MCP Servers
deepcontext
DeepContext is an MCP server that adds symbol-aware semantic search to Claude Code, Codex CLI, and other agents for faster, smarter context on large codebases.
c4-genai-suite
c4 GenAI Suite
claude-code-open
Open source AI coding platform with Web IDE, multi-agent system, 37+ tools, MCP protocol. MIT licensed.
sugar
🍰 Sugar - The autonomous layer for AI coding agents
adal-cli
The self-evolving coding agent that learns from your entire team and codebase. Less syncing. Less waiting. Deliver at the speed of thought.
vsync
Sync MCP servers, Skills, Agents & Commands across Claude Code, Cursor, OpenCode, Codex. One config, all tools.