Code Relay is an MCP-based coding agent integration tool for cross-machine development and verification. It lets AI implement requests on one machine (Dev Host), run real verification on another machine (Target Host), and return results through structured receipts to guide the next round of fix iterations. It supports MCP-compatible coding agents such as Codex, Claude Code, Cursor, and VS Code.
[Purpose] Code Relay is suitable for the following scenarios:
- Different environments: develop on Windows or macOS, and verify on Linux, a private network, or a production-like host.
- Special hardware: send model code to a CUDA/GPU machine, or verify connection locations for cameras, serial devices, sensors, etc.
- Real integration: test payment callbacks, queues, databases, browsers, or internal services that cannot be reproduced locally.
- AI-driven iteration: return specific expected/actual results so the coding agent can fix, republish, and retry.
[Main capabilities]
- Branch-scoped project binding and short-term join links.
- Precisely verify source code commits in isolated worktrees.
- Execute checkpoints through the code-relay-checkpoint GitHub Actions self-hosted runner.
- Allowlisted secure verification commands, with output boundaries and timeout limits.
- Generate structured receipt.json and human-readable receipts, recording passed, failed, or blocked runs.
[Usage] AI client installation: the coding agent can be asked to read and follow the relevant instructions, replacing codex with claude-code, cursor, vscode, or generic, etc. MCP server details: Code Relay is a local MCP server that communicates over stdio. The npm package starts the server. The server implements the MCP JSON-RPC interface directly in Go (without relying on a third-party MCP SDK), supports initialize, ping, tools/list, and tools/call, and exposes the corresponding tools. The implementation and transport loop are located in internal/relay/mcp.go. Workflow: the Dev Host sends a runbook binding the repository and branch through Relay; the Target Host checks the exact source code commit in its real environment and returnsa receipt recording the checkpoint result. Each runbook is bound to a repository, branch, and source code commit. The Target Host acts as a checkpoint, and the generated receipt records passed, failed, or blocked verification for the next round of iteration. Related guides include: AI client installation guide, local desktop installation runbook, deployment and runbook.
[Limitations] The material does not provide pricing, a specific complete list of platform support, detailed runbook contents, or other usage limitations.
