Two production MCP servers, MIT-licensed and pip/npx-installable
Built to make agent-to-backend dispatch a one-config-line operation. Both serve as open-source proof of pattern that any AI agent can integrate with on day one.
atlas-bridge-mcp
Bridges any MCP-compatible AI agent (Claude Desktop, Cursor, Continue, OpenClaw) to a browser-agent / puppeteer HTTP backend. Auto-translates WSL paths to Windows so an agent on Linux can drive a Windows-side automation server transparently. Roughly 80 lines of code that replaces hundreds of lines of per-host plugin code.
Repo: github.com/drelf/atlas-bridge-mcp
claude-bridge-mcp
Bridges MCP agents to a Claude Code CLI subprocess for autonomous coding-task dispatch. Lets one agent (the orchestrator) hand off to another (the implementer) when the task needs filesystem access and multi-step iteration.
Repo: github.com/drelf/claude-bridge-mcp
The pattern
If you're building an AI agent stack and need it to delegate work to specialized backends, these are the simplest possible bridges. One file each, npm-distributable, no framework lock-in. Working installs documented in each README.