Spark Worker
Hand grunt work to your own local AI.
An MCP server that lets a frontier coding agent offload file reading, log digesting and even whole coding tasks to a local model running on owned DGX Spark hardware. Big inputs never enter the expensive agent's context, so sessions last longer.
- MCP tools
- 15
→Reads files and logs locally and returns only the condensed answer
→Delegates whole coding jobs on isolated branches
→One human approval per session, then hands-free
Two ways to read it
In plain English
An MCP server that lets a frontier coding agent offload file reading, log digesting and even whole coding tasks to a local model running on owned DGX Spark hardware. Big inputs never enter the expensive agent's context, so sessions last longer.
For engineers
Python + uv MCP server with 15 tools; delegated coding jobs run on isolated git worktrees behind a machine-wide concurrency cap and red-flag review packets.
What it does.
- Reads files and logs locally and returns only the condensed answer
- Delegates whole coding jobs on isolated branches
- One human approval per session, then hands-free
Built with
- Python
- MCP
- GLM on DGX Spark
- git worktrees
Status: In progress · started October 2026 · private repository